| 1 | |
| 2 |
|
| 3 | |
| 4 | * VMA-specific functions.
|
| 5 |
|
| 6 |
|
| 7 | |
| 8 | * To allow for userland testing we place internal dependencies in
|
| 9 | * vma_internal.h and external VMA API declarations in vma.h.
|
| 10 |
|
| 11 | #include "vma_internal.h"
|
| 12 | #include "vma.h"
|
| 13 |
|
| 14 | struct mmap_state {
|
| 15 | struct mm_struct *mm;
|
| 16 | struct vma_iterator *vmi;
|
| 17 |
|
| 18 | unsigned long addr;
|
| 19 | unsigned long end;
|
| 20 | pgoff_t pgoff;
|
| 21 | pgoff_t anon_pgoff;
|
| 22 | unsigned long pglen;
|
| 23 | union {
|
| 24 | vm_flags_t vm_flags;
|
| 25 | vma_flags_t vma_flags;
|
| 26 | };
|
| 27 | struct file *file;
|
| 28 | pgprot_t page_prot;
|
| 29 |
|
| 30 |
|
| 31 | const struct vm_operations_struct *vm_ops;
|
| 32 | void *vm_private_data;
|
| 33 |
|
| 34 | unsigned long charged;
|
| 35 |
|
| 36 | struct vm_area_struct *prev;
|
| 37 | struct vm_area_struct *next;
|
| 38 |
|
| 39 |
|
| 40 | struct vma_munmap_struct vms;
|
| 41 | struct ma_state mas_detach;
|
| 42 | struct maple_tree mt_detach;
|
| 43 |
|
| 44 |
|
| 45 | bool check_ksm_early :1;
|
| 46 |
|
| 47 | bool file_doesnt_need_get :1;
|
| 48 | };
|
| 49 |
|
| 50 | #define MMAP_STATE(name, mm_, vmi_, addr_, len_, pgoff_, anon_pgoff_, vma_flags_, file_) \
|
| 51 | struct mmap_state name = { \
|
| 52 | .mm = mm_, \
|
| 53 | .vmi = vmi_, \
|
| 54 | .addr = addr_, \
|
| 55 | .end = (addr_) + (len_), \
|
| 56 | .pgoff = pgoff_, \
|
| 57 | .anon_pgoff = anon_pgoff_, \
|
| 58 | .pglen = PHYS_PFN(len_), \
|
| 59 | .vma_flags = vma_flags_, \
|
| 60 | .file = file_, \
|
| 61 | .page_prot = vma_flags_to_page_prot(vma_flags_), \
|
| 62 | }
|
| 63 |
|
| 64 | #define VMG_MMAP_STATE(name, map_, vma_) \
|
| 65 | struct vma_merge_struct name = { \
|
| 66 | .mm = (map_)->mm, \
|
| 67 | .vmi = (map_)->vmi, \
|
| 68 | .start = (map_)->addr, \
|
| 69 | .end = (map_)->end, \
|
| 70 | .vma_flags = (map_)->vma_flags, \
|
| 71 | .pgoff = (map_)->pgoff, \
|
| 72 | .anon_pgoff = (map_)->anon_pgoff, \
|
| 73 | .file = (map_)->file, \
|
| 74 | .prev = (map_)->prev, \
|
| 75 | .middle = vma_, \
|
| 76 | .next = (vma_) ? NULL : (map_)->next, \
|
| 77 | .state = VMA_MERGE_START, \
|
| 78 | }
|
| 79 |
|
| 80 | static void __vma_set_range(struct vm_area_struct *vma, unsigned long start,
|
| 81 | unsigned long end)
|
| 82 | {
|
| 83 | vma->vm_start = start;
|
| 84 | vma->vm_end = end;
|
| 85 | }
|
| 86 |
|
| 87 | static void vma_set_range(struct vm_area_struct *vma, unsigned long start,
|
| 88 | unsigned long end, pgoff_t pgoff, pgoff_t anon_pgoff)
|
| 89 | {
|
| 90 | __vma_set_range(vma, start, end);
|
| 91 | vma_set_pgoff(vma, pgoff);
|
| 92 | vma_set_anon_pgoff(vma, anon_pgoff);
|
| 93 | }
|
| 94 |
|
| 95 |
|
| 96 | static bool vma_is_fork_child(struct vm_area_struct *vma)
|
| 97 | {
|
| 98 | |
| 99 | * The list_is_singular() test is to avoid merging VMA cloned from
|
| 100 | * parents. This can improve scalability caused by the anon_vma root
|
| 101 | * lock.
|
| 102 |
|
| 103 | return vma && vma->anon_vma && !list_is_singular(&vma->anon_vma_chain);
|
| 104 | }
|
| 105 |
|
| 106 | static inline bool is_mergeable_vma(struct vma_merge_struct *vmg, bool merge_next)
|
| 107 | {
|
| 108 | struct vm_area_struct *vma = merge_next ? vmg->next : vmg->prev;
|
| 109 | vma_flags_t diff;
|
| 110 |
|
| 111 | if (!mpol_equal(vmg->policy, vma_policy(vma)))
|
| 112 | return false;
|
| 113 |
|
| 114 | diff = vma_flags_diff_pair(&vma->flags, &vmg->vma_flags);
|
| 115 | vma_flags_clear_mask(&diff, VMA_IGNORE_MERGE_FLAGS);
|
| 116 |
|
| 117 | if (!vma_flags_empty(&diff))
|
| 118 | return false;
|
| 119 | if (vma->vm_file != vmg->file)
|
| 120 | return false;
|
| 121 | if (!is_mergeable_vm_userfaultfd_ctx(vma, vmg->uffd_ctx))
|
| 122 | return false;
|
| 123 | if (!anon_vma_name_eq(anon_vma_name(vma), vmg->anon_name))
|
| 124 | return false;
|
| 125 | return true;
|
| 126 | }
|
| 127 |
|
| 128 | static bool is_mergeable_anon_vma(struct vma_merge_struct *vmg, bool merge_next)
|
| 129 | {
|
| 130 | struct vm_area_struct *tgt = merge_next ? vmg->next : vmg->prev;
|
| 131 | struct vm_area_struct *src = vmg->middle;
|
| 132 | struct anon_vma *tgt_anon = tgt->anon_vma;
|
| 133 | struct anon_vma *src_anon = vmg->anon_vma;
|
| 134 |
|
| 135 | |
| 136 | * We _can_ have !src, vmg->anon_vma via copy_vma(). In this instance we
|
| 137 | * will remove the existing VMA's anon_vma's so there's no scalability
|
| 138 | * concerns.
|
| 139 |
|
| 140 | VM_WARN_ON(src && src_anon != src->anon_vma);
|
| 141 |
|
| 142 |
|
| 143 | if (!tgt_anon && src_anon) {
|
| 144 | struct vm_area_struct *copied_from = vmg->copied_from;
|
| 145 |
|
| 146 | if (vma_is_fork_child(src))
|
| 147 | return false;
|
| 148 | if (vma_is_fork_child(copied_from))
|
| 149 | return false;
|
| 150 |
|
| 151 | return true;
|
| 152 | }
|
| 153 |
|
| 154 | if (tgt_anon && !src_anon)
|
| 155 | return !vma_is_fork_child(tgt);
|
| 156 |
|
| 157 | return src_anon == tgt_anon;
|
| 158 | }
|
| 159 |
|
| 160 | |
| 161 | * init_multi_vma_prep() - Initializer for struct vma_prepare
|
| 162 | * @vp: The vma_prepare struct
|
| 163 | * @vma: The vma that will be altered once locked
|
| 164 | * @vmg: The merge state that will be used to determine adjustment and VMA
|
| 165 | * removal.
|
| 166 |
|
| 167 | static void init_multi_vma_prep(struct vma_prepare *vp,
|
| 168 | struct vm_area_struct *vma,
|
| 169 | struct vma_merge_struct *vmg)
|
| 170 | {
|
| 171 | struct vm_area_struct *adjust;
|
| 172 | struct vm_area_struct **remove = &vp->remove;
|
| 173 |
|
| 174 | memset(vp, 0, sizeof(struct vma_prepare));
|
| 175 | vp->vma = vma;
|
| 176 | vp->anon_vma = vma->anon_vma;
|
| 177 |
|
| 178 | if (vmg && vmg->__remove_middle) {
|
| 179 | *remove = vmg->middle;
|
| 180 | remove = &vp->remove2;
|
| 181 | }
|
| 182 | if (vmg && vmg->__remove_next)
|
| 183 | *remove = vmg->next;
|
| 184 |
|
| 185 | if (vmg && vmg->__adjust_middle_start)
|
| 186 | adjust = vmg->middle;
|
| 187 | else if (vmg && vmg->__adjust_next_start)
|
| 188 | adjust = vmg->next;
|
| 189 | else
|
| 190 | adjust = NULL;
|
| 191 |
|
| 192 | vp->adj_next = adjust;
|
| 193 | if (!vp->anon_vma && adjust)
|
| 194 | vp->anon_vma = adjust->anon_vma;
|
| 195 |
|
| 196 | VM_WARN_ON(vp->anon_vma && adjust && adjust->anon_vma &&
|
| 197 | vp->anon_vma != adjust->anon_vma);
|
| 198 |
|
| 199 | vp->file = vma->vm_file;
|
| 200 | if (vp->file)
|
| 201 | vp->mapping = vma->vm_file->f_mapping;
|
| 202 |
|
| 203 | if (vmg && vmg->skip_vma_uprobe)
|
| 204 | vp->skip_vma_uprobe = true;
|
| 205 | }
|
| 206 |
|
| 207 | |
| 208 | * Does this merge require that adjacent VMAs must have adjacent anonymous page
|
| 209 | * offsets in addition to having adjacent vma->vm_pgoff?
|
| 210 | *
|
| 211 | * This is only required for MAP_PRIVATE-file backed mappings as the page offset
|
| 212 | * for pure anonymous VMAs is equal to the anonymous page offset.
|
| 213 | *
|
| 214 | * Read-only shared mappings (with VMA_SHARED_BIT cleared) are always unfaulted
|
| 215 | * so automatically have correct anonymous page offset (as it is always updated
|
| 216 | * on remap).
|
| 217 | *
|
| 218 | * 'Special' mappings in the sense of VDSO, VVAR etc. have !file but would in
|
| 219 | * any case not be candidates for merge nor be mergeable.
|
| 220 |
|
| 221 | static bool needs_adjacent_anon_pgoff(const struct vma_merge_struct *vmg)
|
| 222 | {
|
| 223 | return vmg->file && vma_flags_is_cow_mapping(&vmg->vma_flags);
|
| 224 | }
|
| 225 |
|
| 226 | |
| 227 | * Return true if we can merge this (vma_flags,anon_vma,file,vm_pgoff)
|
| 228 | * in front of (at a lower virtual address and file offset than) the vma.
|
| 229 | *
|
| 230 | * We cannot merge two vmas if they have differently assigned (non-NULL)
|
| 231 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
|
| 232 | *
|
| 233 | * We don't check here for the merged mmap wrapping around the end of pagecache
|
| 234 | * indices (16TB on ia32) because do_mmap() does not permit mmap's which
|
| 235 | * wrap, nor mmaps which cover the final page at index -1UL.
|
| 236 | *
|
| 237 | * We assume the vma may be removed as part of the merge.
|
| 238 |
|
| 239 | static bool can_vma_merge_before(struct vma_merge_struct *vmg)
|
| 240 | {
|
| 241 | if (!is_mergeable_vma(vmg, true))
|
| 242 | return false;
|
| 243 | if (!is_mergeable_anon_vma(vmg, true))
|
| 244 | return false;
|
| 245 | if (vmg_end_pgoff(vmg) != vma_start_pgoff(vmg->next))
|
| 246 | return false;
|
| 247 | if (needs_adjacent_anon_pgoff(vmg) &&
|
| 248 | vmg_end_anon_pgoff(vmg) != vma_start_anon_pgoff(vmg->next))
|
| 249 | return false;
|
| 250 | return true;
|
| 251 | }
|
| 252 |
|
| 253 | |
| 254 | * Return true if we can merge this (vma_flags,anon_vma,file,vm_pgoff)
|
| 255 | * beyond (at a higher virtual address and file offset than) the vma.
|
| 256 | *
|
| 257 | * We cannot merge two vmas if they have differently assigned (non-NULL)
|
| 258 | * anon_vmas, nor if same anon_vma is assigned but offsets incompatible.
|
| 259 | *
|
| 260 | * We assume that vma is not removed as part of the merge.
|
| 261 |
|
| 262 | static bool can_vma_merge_after(struct vma_merge_struct *vmg)
|
| 263 | {
|
| 264 | if (!is_mergeable_vma(vmg, false))
|
| 265 | return false;
|
| 266 | if (!is_mergeable_anon_vma(vmg, false))
|
| 267 | return false;
|
| 268 | if (vma_end_pgoff(vmg->prev) != vmg_start_pgoff(vmg))
|
| 269 | return false;
|
| 270 | if (needs_adjacent_anon_pgoff(vmg) &&
|
| 271 | vma_end_anon_pgoff(vmg->prev) != vmg_start_anon_pgoff(vmg))
|
| 272 | return false;
|
| 273 | return true;
|
| 274 | }
|
| 275 |
|
| 276 | static void __vma_link_file(struct vm_area_struct *vma,
|
| 277 | struct address_space *mapping)
|
| 278 | {
|
| 279 | if (vma_is_shared_maywrite(vma))
|
| 280 | mapping_allow_writable(mapping);
|
| 281 |
|
| 282 | flush_dcache_mmap_lock(mapping);
|
| 283 | mapping_rmap_tree_insert(vma, mapping);
|
| 284 | flush_dcache_mmap_unlock(mapping);
|
| 285 | }
|
| 286 |
|
| 287 | |
| 288 | * Requires inode->i_mapping->i_mmap_rwsem
|
| 289 |
|
| 290 | static void __remove_shared_vm_struct(struct vm_area_struct *vma,
|
| 291 | struct address_space *mapping)
|
| 292 | {
|
| 293 | if (vma_is_shared_maywrite(vma))
|
| 294 | mapping_unmap_writable(mapping);
|
| 295 |
|
| 296 | flush_dcache_mmap_lock(mapping);
|
| 297 | mapping_rmap_tree_remove(vma, mapping);
|
| 298 | flush_dcache_mmap_unlock(mapping);
|
| 299 | }
|
| 300 |
|
| 301 | |
| 302 | * vma has some anon_vma assigned, and is already inserted on that
|
| 303 | * anon_vma's interval trees.
|
| 304 | *
|
| 305 | * Before updating the vma's vm_start / vm_end / vm_pgoff fields, the
|
| 306 | * vma must be removed from the anon_vma's interval trees using
|
| 307 | * anon_rmap_tree_pre_update_vma().
|
| 308 | *
|
| 309 | * After the update, the vma will be reinserted using
|
| 310 | * anon_rmap_tree_post_update_vma().
|
| 311 | *
|
| 312 | * The entire update must be protected by exclusive mmap_lock and by
|
| 313 | * the root anon_vma's mutex.
|
| 314 |
|
| 315 | static void
|
| 316 | anon_rmap_tree_pre_update_vma(struct vm_area_struct *vma)
|
| 317 | {
|
| 318 | struct anon_vma_chain *avc;
|
| 319 |
|
| 320 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
|
| 321 | anon_rmap_tree_remove(avc, avc->anon_vma);
|
| 322 | }
|
| 323 |
|
| 324 | static void
|
| 325 | anon_rmap_tree_post_update_vma(struct vm_area_struct *vma)
|
| 326 | {
|
| 327 | struct anon_vma_chain *avc;
|
| 328 |
|
| 329 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
|
| 330 | anon_rmap_tree_insert(avc, avc->anon_vma);
|
| 331 | }
|
| 332 |
|
| 333 | |
| 334 | * vma_prepare() - Helper function for handling locking VMAs prior to altering
|
| 335 | * @vp: The initialized vma_prepare struct
|
| 336 |
|
| 337 | static void vma_prepare(struct vma_prepare *vp)
|
| 338 | {
|
| 339 | if (vp->file) {
|
| 340 | uprobe_munmap(vp->vma, vp->vma->vm_start, vp->vma->vm_end);
|
| 341 |
|
| 342 | if (vp->adj_next)
|
| 343 | uprobe_munmap(vp->adj_next, vp->adj_next->vm_start,
|
| 344 | vp->adj_next->vm_end);
|
| 345 |
|
| 346 | i_mmap_lock_write(vp->mapping);
|
| 347 | if (vp->insert && vp->insert->vm_file) {
|
| 348 | |
| 349 | * Put into interval tree now, so instantiated pages
|
| 350 | * are visible to arm/parisc __flush_dcache_page
|
| 351 | * throughout; but we cannot insert into address
|
| 352 | * space until vma start or end is updated.
|
| 353 |
|
| 354 | __vma_link_file(vp->insert,
|
| 355 | vp->insert->vm_file->f_mapping);
|
| 356 | }
|
| 357 | }
|
| 358 |
|
| 359 | if (vp->anon_vma) {
|
| 360 | anon_vma_lock_write(vp->anon_vma);
|
| 361 | anon_rmap_tree_pre_update_vma(vp->vma);
|
| 362 | if (vp->adj_next)
|
| 363 | anon_rmap_tree_pre_update_vma(vp->adj_next);
|
| 364 | }
|
| 365 |
|
| 366 | if (vp->file) {
|
| 367 | flush_dcache_mmap_lock(vp->mapping);
|
| 368 | mapping_rmap_tree_remove(vp->vma, vp->mapping);
|
| 369 | if (vp->adj_next)
|
| 370 | mapping_rmap_tree_remove(vp->adj_next, vp->mapping);
|
| 371 | }
|
| 372 |
|
| 373 | }
|
| 374 |
|
| 375 | |
| 376 | * vma_complete- Helper function for handling the unlocking after altering VMAs,
|
| 377 | * or for inserting a VMA.
|
| 378 | *
|
| 379 | * @vp: The vma_prepare struct
|
| 380 | * @vmi: The vma iterator
|
| 381 | * @mm: The mm_struct
|
| 382 |
|
| 383 | static void vma_complete(struct vma_prepare *vp, struct vma_iterator *vmi,
|
| 384 | struct mm_struct *mm)
|
| 385 | {
|
| 386 | if (vp->file) {
|
| 387 | if (vp->adj_next)
|
| 388 | mapping_rmap_tree_insert(vp->adj_next, vp->mapping);
|
| 389 | mapping_rmap_tree_insert(vp->vma, vp->mapping);
|
| 390 | flush_dcache_mmap_unlock(vp->mapping);
|
| 391 | }
|
| 392 |
|
| 393 | if (vp->remove && vp->file) {
|
| 394 | __remove_shared_vm_struct(vp->remove, vp->mapping);
|
| 395 | if (vp->remove2)
|
| 396 | __remove_shared_vm_struct(vp->remove2, vp->mapping);
|
| 397 | } else if (vp->insert) {
|
| 398 | |
| 399 | * split_vma has split insert from vma, and needs
|
| 400 | * us to insert it before dropping the locks
|
| 401 | * (it may either follow vma or precede it).
|
| 402 |
|
| 403 | vma_iter_store_new(vmi, vp->insert);
|
| 404 | mm->map_count++;
|
| 405 | }
|
| 406 |
|
| 407 | if (vp->anon_vma) {
|
| 408 | anon_rmap_tree_post_update_vma(vp->vma);
|
| 409 | if (vp->adj_next)
|
| 410 | anon_rmap_tree_post_update_vma(vp->adj_next);
|
| 411 | anon_vma_unlock_write(vp->anon_vma);
|
| 412 | }
|
| 413 |
|
| 414 | if (vp->file) {
|
| 415 | i_mmap_unlock_write(vp->mapping);
|
| 416 |
|
| 417 | if (!vp->skip_vma_uprobe) {
|
| 418 | uprobe_mmap(vp->vma);
|
| 419 |
|
| 420 | if (vp->adj_next)
|
| 421 | uprobe_mmap(vp->adj_next);
|
| 422 | }
|
| 423 | }
|
| 424 |
|
| 425 | if (vp->remove) {
|
| 426 | again:
|
| 427 | vma_mark_detached(vp->remove);
|
| 428 | if (vp->file) {
|
| 429 | uprobe_munmap(vp->remove, vp->remove->vm_start,
|
| 430 | vp->remove->vm_end);
|
| 431 | fput(vp->file);
|
| 432 | }
|
| 433 | if (vp->remove->anon_vma)
|
| 434 | unlink_anon_vmas(vp->remove);
|
| 435 | mm->map_count--;
|
| 436 | mpol_put(vma_policy(vp->remove));
|
| 437 | if (!vp->remove2)
|
| 438 | WARN_ON_ONCE(vp->vma->vm_end < vp->remove->vm_end);
|
| 439 | vm_area_free(vp->remove);
|
| 440 |
|
| 441 | |
| 442 | * In mprotect's case 6 (see comments on vma_merge),
|
| 443 | * we are removing both mid and next vmas
|
| 444 |
|
| 445 | if (vp->remove2) {
|
| 446 | vp->remove = vp->remove2;
|
| 447 | vp->remove2 = NULL;
|
| 448 | goto again;
|
| 449 | }
|
| 450 | }
|
| 451 | if (vp->insert && vp->file)
|
| 452 | uprobe_mmap(vp->insert);
|
| 453 | }
|
| 454 |
|
| 455 | |
| 456 | * init_vma_prep() - Initializer wrapper for vma_prepare struct
|
| 457 | * @vp: The vma_prepare struct
|
| 458 | * @vma: The vma that will be altered once locked
|
| 459 |
|
| 460 | static void init_vma_prep(struct vma_prepare *vp, struct vm_area_struct *vma)
|
| 461 | {
|
| 462 | init_multi_vma_prep(vp, vma, NULL);
|
| 463 | }
|
| 464 |
|
| 465 | |
| 466 | * Can the proposed VMA be merged with the left (previous) VMA taking into
|
| 467 | * account the start position of the proposed range.
|
| 468 |
|
| 469 | static bool can_vma_merge_left(struct vma_merge_struct *vmg)
|
| 470 |
|
| 471 | {
|
| 472 | return vmg->prev && vmg->prev->vm_end == vmg->start &&
|
| 473 | can_vma_merge_after(vmg);
|
| 474 | }
|
| 475 |
|
| 476 | |
| 477 | * Can the proposed VMA be merged with the right (next) VMA taking into
|
| 478 | * account the end position of the proposed range.
|
| 479 | *
|
| 480 | * In addition, if we can merge with the left VMA, ensure that left and right
|
| 481 | * anon_vma's are also compatible.
|
| 482 |
|
| 483 | static bool can_vma_merge_right(struct vma_merge_struct *vmg,
|
| 484 | bool can_merge_left)
|
| 485 | {
|
| 486 | struct vm_area_struct *next = vmg->next;
|
| 487 | struct vm_area_struct *prev;
|
| 488 |
|
| 489 | if (!next || vmg->end != next->vm_start || !can_vma_merge_before(vmg))
|
| 490 | return false;
|
| 491 |
|
| 492 | if (!can_merge_left)
|
| 493 | return true;
|
| 494 |
|
| 495 | |
| 496 | * If we can merge with prev (left) and next (right), indicating that
|
| 497 | * each VMA's anon_vma is compatible with the proposed anon_vma, this
|
| 498 | * does not mean prev and next are compatible with EACH OTHER.
|
| 499 | *
|
| 500 | * We therefore check this in addition to mergeability to either side.
|
| 501 |
|
| 502 | prev = vmg->prev;
|
| 503 | return !prev->anon_vma || !next->anon_vma ||
|
| 504 | prev->anon_vma == next->anon_vma;
|
| 505 | }
|
| 506 |
|
| 507 | |
| 508 | * Close a vm structure and free it.
|
| 509 |
|
| 510 | void remove_vma(struct vm_area_struct *vma)
|
| 511 | {
|
| 512 | might_sleep();
|
| 513 | vma_close(vma);
|
| 514 | if (vma->vm_file)
|
| 515 | fput(vma->vm_file);
|
| 516 | mpol_put(vma_policy(vma));
|
| 517 | vm_area_free(vma);
|
| 518 | }
|
| 519 |
|
| 520 | |
| 521 | * Get rid of page table information in the indicated region.
|
| 522 | *
|
| 523 | * Called with the mm semaphore held.
|
| 524 |
|
| 525 | void unmap_region(struct unmap_desc *unmap)
|
| 526 | {
|
| 527 | struct mm_struct *mm = unmap->first->vm_mm;
|
| 528 | struct mmu_gather tlb;
|
| 529 |
|
| 530 | tlb_gather_mmu(&tlb, mm);
|
| 531 | update_hiwater_rss(mm);
|
| 532 | unmap_vmas(&tlb, unmap);
|
| 533 | mas_set(unmap->mas, unmap->tree_reset);
|
| 534 | free_pgtables(&tlb, unmap);
|
| 535 | tlb_finish_mmu(&tlb);
|
| 536 | }
|
| 537 |
|
| 538 | |
| 539 | * __split_vma() bypasses sysctl_max_map_count checking. We use this where it
|
| 540 | * has already been checked or doesn't make sense to fail.
|
| 541 | * VMA Iterator will point to the original VMA.
|
| 542 |
|
| 543 | static __must_check int
|
| 544 | __split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
|
| 545 | unsigned long addr, int new_below)
|
| 546 | {
|
| 547 | struct vma_prepare vp;
|
| 548 | struct vm_area_struct *new;
|
| 549 | int err;
|
| 550 |
|
| 551 | WARN_ON(vma->vm_start >= addr);
|
| 552 | WARN_ON(vma->vm_end <= addr);
|
| 553 |
|
| 554 | if (vma->vm_ops && vma->vm_ops->may_split) {
|
| 555 | err = vma->vm_ops->may_split(vma, addr);
|
| 556 | if (err)
|
| 557 | return err;
|
| 558 | }
|
| 559 |
|
| 560 | new = vm_area_dup(vma);
|
| 561 | if (!new)
|
| 562 | return -ENOMEM;
|
| 563 |
|
| 564 | if (new_below) {
|
| 565 | new->vm_end = addr;
|
| 566 | } else {
|
| 567 | new->vm_start = addr;
|
| 568 | vma_add_pgoff(new, linear_page_delta(vma, addr));
|
| 569 | }
|
| 570 |
|
| 571 | err = -ENOMEM;
|
| 572 | vma_iter_config(vmi, new->vm_start, new->vm_end);
|
| 573 | if (vma_iter_prealloc(vmi, new))
|
| 574 | goto out_free_vma;
|
| 575 |
|
| 576 | err = vma_dup_policy(vma, new);
|
| 577 | if (err)
|
| 578 | goto out_free_vmi;
|
| 579 |
|
| 580 | err = anon_vma_clone(new, vma, VMA_OP_SPLIT);
|
| 581 | if (err)
|
| 582 | goto out_free_mpol;
|
| 583 |
|
| 584 | if (new->vm_file)
|
| 585 | get_file(new->vm_file);
|
| 586 |
|
| 587 | if (new->vm_ops && new->vm_ops->open)
|
| 588 | new->vm_ops->open(new);
|
| 589 |
|
| 590 | vma_start_write(vma);
|
| 591 | vma_start_write(new);
|
| 592 |
|
| 593 | init_vma_prep(&vp, vma);
|
| 594 | vp.insert = new;
|
| 595 | vma_prepare(&vp);
|
| 596 |
|
| 597 | |
| 598 | * Get rid of huge pages and shared page tables straddling the split
|
| 599 | * boundary.
|
| 600 |
|
| 601 | vma_adjust_trans_huge(vma, vma->vm_start, addr, NULL);
|
| 602 | if (is_vm_hugetlb_page(vma))
|
| 603 | hugetlb_split(vma, addr);
|
| 604 |
|
| 605 | if (new_below) {
|
| 606 | vma->vm_start = addr;
|
| 607 | vma_add_pgoff(vma, linear_page_delta(new, addr));
|
| 608 | } else {
|
| 609 | vma->vm_end = addr;
|
| 610 | }
|
| 611 |
|
| 612 |
|
| 613 | vma_complete(&vp, vmi, vma->vm_mm);
|
| 614 | validate_mm(vma->vm_mm);
|
| 615 |
|
| 616 |
|
| 617 | if (new_below)
|
| 618 | vma_next(vmi);
|
| 619 | else
|
| 620 | vma_prev(vmi);
|
| 621 |
|
| 622 | return 0;
|
| 623 |
|
| 624 | out_free_mpol:
|
| 625 | mpol_put(vma_policy(new));
|
| 626 | out_free_vmi:
|
| 627 | vma_iter_free(vmi);
|
| 628 | out_free_vma:
|
| 629 | vm_area_free(new);
|
| 630 | return err;
|
| 631 | }
|
| 632 |
|
| 633 | |
| 634 | * Split a vma into two pieces at address 'addr', a new vma is allocated
|
| 635 | * either for the first part or the tail.
|
| 636 |
|
| 637 | int split_vma(struct vma_iterator *vmi, struct vm_area_struct *vma,
|
| 638 | unsigned long addr, int new_below)
|
| 639 | {
|
| 640 | if (vma->vm_mm->map_count >= get_sysctl_max_map_count())
|
| 641 | return -ENOMEM;
|
| 642 |
|
| 643 | return __split_vma(vmi, vma, addr, new_below);
|
| 644 | }
|
| 645 |
|
| 646 | |
| 647 | * dup_anon_vma() - Helper function to duplicate anon_vma on VMA merge in the
|
| 648 | * instance that the destination VMA has no anon_vma but the source does.
|
| 649 | *
|
| 650 | * @dst: The destination VMA
|
| 651 | * @src: The source VMA
|
| 652 | * @dup: Pointer to the destination VMA when successful.
|
| 653 | *
|
| 654 | * Returns: 0 on success.
|
| 655 |
|
| 656 | static int dup_anon_vma(struct vm_area_struct *dst,
|
| 657 | struct vm_area_struct *src, struct vm_area_struct **dup)
|
| 658 | {
|
| 659 | |
| 660 | * There are three cases to consider for correctly propagating
|
| 661 | * anon_vma's on merge.
|
| 662 | *
|
| 663 | * The first is trivial - neither VMA has anon_vma, we need not do
|
| 664 | * anything.
|
| 665 | *
|
| 666 | * The second where both have anon_vma is also a no-op, as they must
|
| 667 | * then be the same, so there is simply nothing to copy.
|
| 668 | *
|
| 669 | * Here we cover the third - if the destination VMA has no anon_vma,
|
| 670 | * that is it is unfaulted, we need to ensure that the newly merged
|
| 671 | * range is referenced by the anon_vma's of the source.
|
| 672 |
|
| 673 | if (src->anon_vma && !dst->anon_vma) {
|
| 674 | int ret;
|
| 675 |
|
| 676 | vma_assert_write_locked(dst);
|
| 677 | dst->anon_vma = src->anon_vma;
|
| 678 | ret = anon_vma_clone(dst, src, VMA_OP_MERGE_UNFAULTED);
|
| 679 | if (ret)
|
| 680 | return ret;
|
| 681 |
|
| 682 | *dup = dst;
|
| 683 | }
|
| 684 |
|
| 685 | return 0;
|
| 686 | }
|
| 687 |
|
| 688 | #ifdef CONFIG_DEBUG_VM_MAPLE_TREE
|
| 689 | void validate_mm(struct mm_struct *mm)
|
| 690 | {
|
| 691 | int bug = 0;
|
| 692 | int i = 0;
|
| 693 | struct vm_area_struct *vma;
|
| 694 | VMA_ITERATOR(vmi, mm, 0);
|
| 695 |
|
| 696 | mt_validate(&mm->mm_mt);
|
| 697 | for_each_vma(vmi, vma) {
|
| 698 | #ifdef CONFIG_DEBUG_VM_RB
|
| 699 | struct anon_vma *anon_vma = vma->anon_vma;
|
| 700 | struct anon_vma_chain *avc;
|
| 701 | #endif
|
| 702 | unsigned long vmi_start, vmi_end;
|
| 703 | bool warn = 0;
|
| 704 |
|
| 705 | vmi_start = vma_iter_addr(&vmi);
|
| 706 | vmi_end = vma_iter_end(&vmi);
|
| 707 | if (VM_WARN_ON_ONCE_MM(vma->vm_end != vmi_end, mm))
|
| 708 | warn = 1;
|
| 709 |
|
| 710 | if (VM_WARN_ON_ONCE_MM(vma->vm_start != vmi_start, mm))
|
| 711 | warn = 1;
|
| 712 |
|
| 713 | if (warn) {
|
| 714 | pr_emerg("issue in %s\n", current->comm);
|
| 715 | dump_stack();
|
| 716 | dump_vma(vma);
|
| 717 | pr_emerg("tree range: %px start %lx end %lx\n", vma,
|
| 718 | vmi_start, vmi_end - 1);
|
| 719 | vma_iter_dump_tree(&vmi);
|
| 720 | }
|
| 721 |
|
| 722 | #ifdef CONFIG_DEBUG_VM_RB
|
| 723 | if (anon_vma) {
|
| 724 | anon_vma_lock_read(anon_vma);
|
| 725 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
|
| 726 | anon_rmap_tree_verify(avc);
|
| 727 | anon_vma_unlock_read(anon_vma);
|
| 728 | }
|
| 729 | #endif
|
| 730 |
|
| 731 | if (++i > mm->map_count + 10) {
|
| 732 | i = -1;
|
| 733 | break;
|
| 734 | }
|
| 735 | }
|
| 736 | if (i != mm->map_count) {
|
| 737 | pr_emerg("map_count %d vma iterator %d\n", mm->map_count, i);
|
| 738 | bug = 1;
|
| 739 | }
|
| 740 | VM_BUG_ON_MM(bug, mm);
|
| 741 | }
|
| 742 | #endif
|
| 743 |
|
| 744 | |
| 745 | * Based on the vmg flag indicating whether we need to adjust the vm_start field
|
| 746 | * for the middle or next VMA, we calculate what the range of the newly adjusted
|
| 747 | * VMA ought to be, and set the VMA's range accordingly.
|
| 748 |
|
| 749 | static void vmg_adjust_set_range(struct vma_merge_struct *vmg)
|
| 750 | {
|
| 751 | if (vmg->__adjust_middle_start) {
|
| 752 | |
| 753 | * vmg->start vmg->end
|
| 754 | * | |
|
| 755 | * v merge v
|
| 756 | * <------------->
|
| 757 | * delta
|
| 758 | * <------>
|
| 759 | * |------|----------------|
|
| 760 | * | prev | middle |
|
| 761 | * |------|----------------|
|
| 762 | * ^
|
| 763 | * |
|
| 764 | * middle->vm_start
|
| 765 |
|
| 766 | struct vm_area_struct *middle = vmg->middle;
|
| 767 | const unsigned long delta = vmg->end - middle->vm_start;
|
| 768 |
|
| 769 | __vma_set_range(middle, vmg->end, middle->vm_end);
|
| 770 | vma_add_pgoff(middle, delta >> PAGE_SHIFT);
|
| 771 | } else if (vmg->__adjust_next_start) {
|
| 772 | |
| 773 | * Originally:
|
| 774 | *
|
| 775 | * vmg->start vmg->end
|
| 776 | * | |
|
| 777 | * v merge v
|
| 778 | * <------------>
|
| 779 | * . .
|
| 780 | * merge_existing_range() updates to:
|
| 781 | * . .
|
| 782 | * vmg->start vmg->end .
|
| 783 | * | | .
|
| 784 | * v retain v .
|
| 785 | * <----------> .
|
| 786 | * delta .
|
| 787 | * <-----> .
|
| 788 | * |----------------|------|
|
| 789 | * | middle | next |
|
| 790 | * |----------------|------|
|
| 791 | * ^
|
| 792 | * |
|
| 793 | * next->vm_start
|
| 794 |
|
| 795 | struct vm_area_struct *next = vmg->next;
|
| 796 | const unsigned long delta = next->vm_start - vmg->end;
|
| 797 |
|
| 798 | __vma_set_range(next, vmg->end, next->vm_end);
|
| 799 | vma_sub_pgoff(next, delta >> PAGE_SHIFT);
|
| 800 | }
|
| 801 | }
|
| 802 |
|
| 803 | |
| 804 | * Actually perform the VMA merge operation.
|
| 805 | *
|
| 806 | * IMPORTANT: We guarantee that, should vmg->give_up_on_oom is set, to not
|
| 807 | * modify any VMAs or cause inconsistent state should an OOM condition arise.
|
| 808 | *
|
| 809 | * Returns 0 on success, or an error value on failure.
|
| 810 |
|
| 811 | static int commit_merge(struct vma_merge_struct *vmg)
|
| 812 | {
|
| 813 | struct vm_area_struct *vma;
|
| 814 | struct vma_prepare vp;
|
| 815 |
|
| 816 | if (vmg->__adjust_next_start) {
|
| 817 |
|
| 818 | vma = vmg->middle;
|
| 819 | vma_iter_config(vmg->vmi, vmg->end, vmg->next->vm_end);
|
| 820 | } else {
|
| 821 | vma = vmg->target;
|
| 822 |
|
| 823 | vma_iter_config(vmg->vmi, vmg->start, vmg->end);
|
| 824 | }
|
| 825 |
|
| 826 | init_multi_vma_prep(&vp, vma, vmg);
|
| 827 |
|
| 828 | |
| 829 | * If vmg->give_up_on_oom is set, we're safe, because we don't actually
|
| 830 | * manipulate any VMAs until we succeed at preallocation.
|
| 831 | *
|
| 832 | * Past this point, we will not return an error.
|
| 833 |
|
| 834 | if (vma_iter_prealloc(vmg->vmi, vma))
|
| 835 | return -ENOMEM;
|
| 836 |
|
| 837 | vma_prepare(&vp);
|
| 838 | |
| 839 | * THP pages may need to do additional splits if we increase
|
| 840 | * middle->vm_start.
|
| 841 |
|
| 842 | vma_adjust_trans_huge(vma, vmg->start, vmg->end,
|
| 843 | vmg->__adjust_middle_start ? vmg->middle : NULL);
|
| 844 | vma_set_range(vma, vmg->start, vmg->end, vmg_start_pgoff(vmg),
|
| 845 | vmg_start_anon_pgoff(vmg));
|
| 846 | vmg_adjust_set_range(vmg);
|
| 847 | vma_iter_store_overwrite(vmg->vmi, vmg->target);
|
| 848 |
|
| 849 | vma_complete(&vp, vmg->vmi, vma->vm_mm);
|
| 850 |
|
| 851 | return 0;
|
| 852 | }
|
| 853 |
|
| 854 |
|
| 855 | static bool can_merge_remove_vma(struct vm_area_struct *vma)
|
| 856 | {
|
| 857 | return !vma->vm_ops || !vma->vm_ops->close;
|
| 858 | }
|
| 859 |
|
| 860 | |
| 861 | * vma_merge_existing_range - Attempt to merge VMAs based on a VMA having its
|
| 862 | * attributes modified.
|
| 863 | *
|
| 864 | * @vmg: Describes the modifications being made to a VMA and associated
|
| 865 | * metadata.
|
| 866 | *
|
| 867 | * When the attributes of a range within a VMA change, then it might be possible
|
| 868 | * for immediately adjacent VMAs to be merged into that VMA due to having
|
| 869 | * identical properties.
|
| 870 | *
|
| 871 | * This function checks for the existence of any such mergeable VMAs and updates
|
| 872 | * the maple tree describing the @vmg->middle->vm_mm address space to account
|
| 873 | * for this, as well as any VMAs shrunk/expanded/deleted as a result of this
|
| 874 | * merge.
|
| 875 | *
|
| 876 | * As part of this operation, if a merge occurs, the @vmg object will have its
|
| 877 | * vma, start, end, and pgoff fields modified to execute the merge. Subsequent
|
| 878 | * calls to this function should reset these fields.
|
| 879 | *
|
| 880 | * Returns: The merged VMA if merge succeeds, or NULL otherwise.
|
| 881 | *
|
| 882 | * ASSUMPTIONS:
|
| 883 | * - The caller must assign the VMA to be modified to @vmg->middle.
|
| 884 | * - The caller must have set @vmg->prev to the previous VMA, if there is one.
|
| 885 | * - The caller must not set @vmg->next, as we determine this.
|
| 886 | * - The caller must hold a WRITE lock on the mm_struct->mmap_lock.
|
| 887 | * - vmi must be positioned within [@vmg->middle->vm_start, @vmg->middle->vm_end).
|
| 888 |
|
| 889 | static __must_check struct vm_area_struct *vma_merge_existing_range(
|
| 890 | struct vma_merge_struct *vmg)
|
| 891 | {
|
| 892 | vma_flags_t sticky_flags = vma_flags_and_mask(&vmg->vma_flags,
|
| 893 | VMA_STICKY_FLAGS);
|
| 894 | struct vm_area_struct *middle = vmg->middle;
|
| 895 | struct vm_area_struct *prev = vmg->prev;
|
| 896 | struct vm_area_struct *next;
|
| 897 | struct vm_area_struct *anon_dup = NULL;
|
| 898 | unsigned long start = vmg->start;
|
| 899 | unsigned long end = vmg->end;
|
| 900 | bool left_side = middle && start == middle->vm_start;
|
| 901 | bool right_side = middle && end == middle->vm_end;
|
| 902 | int err = 0;
|
| 903 | bool merge_left, merge_right, merge_both;
|
| 904 |
|
| 905 | mmap_assert_write_locked(vmg->mm);
|
| 906 | VM_WARN_ON_VMG(!middle, vmg);
|
| 907 | VM_WARN_ON_VMG(vmg->next, vmg);
|
| 908 | VM_WARN_ON_VMG(prev && start <= prev->vm_start, vmg);
|
| 909 | VM_WARN_ON_VMG(start >= end, vmg);
|
| 910 |
|
| 911 | |
| 912 | * If middle == prev, then we are offset into a VMA. Otherwise, if we are
|
| 913 | * not, we must span a portion of the VMA.
|
| 914 |
|
| 915 | VM_WARN_ON_VMG(middle &&
|
| 916 | ((middle != prev && vmg->start != middle->vm_start) ||
|
| 917 | vmg->end > middle->vm_end), vmg);
|
| 918 |
|
| 919 | VM_WARN_ON_VMG(middle &&
|
| 920 | !(vma_iter_addr(vmg->vmi) >= middle->vm_start &&
|
| 921 | vma_iter_addr(vmg->vmi) < middle->vm_end), vmg);
|
| 922 |
|
| 923 | VM_WARN_ON_VMG(vmg->copied_from, vmg);
|
| 924 |
|
| 925 | vmg->state = VMA_MERGE_NOMERGE;
|
| 926 |
|
| 927 | |
| 928 | * If a special mapping or if the range being modified is neither at the
|
| 929 | * furthermost left or right side of the VMA, then we have no chance of
|
| 930 | * merging and should abort.
|
| 931 |
|
| 932 | if (vma_flags_test_any_mask(&vmg->vma_flags, VMA_SPECIAL_FLAGS) ||
|
| 933 | (!left_side && !right_side))
|
| 934 | return NULL;
|
| 935 |
|
| 936 | if (left_side)
|
| 937 | merge_left = can_vma_merge_left(vmg);
|
| 938 | else
|
| 939 | merge_left = false;
|
| 940 |
|
| 941 | if (right_side) {
|
| 942 | next = vmg->next = vma_iter_next_range(vmg->vmi);
|
| 943 | vma_iter_prev_range(vmg->vmi);
|
| 944 |
|
| 945 | merge_right = can_vma_merge_right(vmg, merge_left);
|
| 946 | } else {
|
| 947 | merge_right = false;
|
| 948 | next = NULL;
|
| 949 | }
|
| 950 |
|
| 951 | if (merge_left)
|
| 952 | vma_prev(vmg->vmi);
|
| 953 | else if (!merge_right)
|
| 954 | return NULL;
|
| 955 |
|
| 956 | merge_both = merge_left && merge_right;
|
| 957 |
|
| 958 | vmg->__remove_middle = left_side && right_side;
|
| 959 |
|
| 960 | |
| 961 | * If we need to remove middle in its entirety but are unable to do so,
|
| 962 | * we have no sensible recourse but to abort the merge.
|
| 963 |
|
| 964 | if (vmg->__remove_middle && !can_merge_remove_vma(middle))
|
| 965 | return NULL;
|
| 966 |
|
| 967 | |
| 968 | * If we merge both VMAs, then next is also deleted. This implies
|
| 969 | * merge_will_delete_vma also.
|
| 970 |
|
| 971 | vmg->__remove_next = merge_both;
|
| 972 |
|
| 973 | |
| 974 | * If we cannot delete next, then we can reduce the operation to merging
|
| 975 | * prev and middle (thereby deleting middle).
|
| 976 |
|
| 977 | if (vmg->__remove_next && !can_merge_remove_vma(next)) {
|
| 978 | vmg->__remove_next = false;
|
| 979 | merge_right = false;
|
| 980 | merge_both = false;
|
| 981 | }
|
| 982 |
|
| 983 |
|
| 984 | vma_start_write(middle);
|
| 985 |
|
| 986 | if (merge_right) {
|
| 987 | vma_flags_t next_sticky;
|
| 988 |
|
| 989 | vma_start_write(next);
|
| 990 | vmg->target = next;
|
| 991 | next_sticky = vma_flags_and_mask(&next->flags, VMA_STICKY_FLAGS);
|
| 992 | vma_flags_set_mask(&sticky_flags, next_sticky);
|
| 993 | }
|
| 994 |
|
| 995 | if (merge_left) {
|
| 996 | vma_flags_t prev_sticky;
|
| 997 |
|
| 998 | vma_start_write(prev);
|
| 999 | vmg->target = prev;
|
| 1000 |
|
| 1001 | prev_sticky = vma_flags_and_mask(&prev->flags, VMA_STICKY_FLAGS);
|
| 1002 | vma_flags_set_mask(&sticky_flags, prev_sticky);
|
| 1003 | }
|
| 1004 |
|
| 1005 | if (merge_both) {
|
| 1006 | |
| 1007 | * |<-------------------->|
|
| 1008 | * |-------********-------|
|
| 1009 | * prev middle next
|
| 1010 | * extend delete delete
|
| 1011 |
|
| 1012 | vmg->start = prev->vm_start;
|
| 1013 | vmg->end = next->vm_end;
|
| 1014 | vmg->pgoff = vma_start_pgoff(prev);
|
| 1015 | vmg->anon_pgoff = vma_start_anon_pgoff(prev);
|
| 1016 |
|
| 1017 | |
| 1018 | * We already ensured anon_vma compatibility above, so now it's
|
| 1019 | * simply a case of, if prev has no anon_vma object, which of
|
| 1020 | * next or middle contains the anon_vma we must duplicate.
|
| 1021 |
|
| 1022 | err = dup_anon_vma(prev, next->anon_vma ? next : middle,
|
| 1023 | &anon_dup);
|
| 1024 | } else if (merge_left) {
|
| 1025 | |
| 1026 | * |<------------>| OR
|
| 1027 | * |<----------------->|
|
| 1028 | * |-------*************
|
| 1029 | * prev middle
|
| 1030 | * extend shrink/delete
|
| 1031 |
|
| 1032 | vmg->start = prev->vm_start;
|
| 1033 | vmg->pgoff = vma_start_pgoff(prev);
|
| 1034 | vmg->anon_pgoff = vma_start_anon_pgoff(prev);
|
| 1035 |
|
| 1036 | if (!vmg->__remove_middle)
|
| 1037 | vmg->__adjust_middle_start = true;
|
| 1038 |
|
| 1039 | err = dup_anon_vma(prev, middle, &anon_dup);
|
| 1040 | } else {
|
| 1041 | |
| 1042 | * |<------------->| OR
|
| 1043 | * |<----------------->|
|
| 1044 | * *************-------|
|
| 1045 | * middle next
|
| 1046 | * shrink/delete extend
|
| 1047 |
|
| 1048 | const pgoff_t pglen = vmg_pages(vmg);
|
| 1049 |
|
| 1050 | VM_WARN_ON_VMG(!merge_right, vmg);
|
| 1051 |
|
| 1052 | VM_WARN_ON_VMG(vmg->start > middle->vm_start && prev && middle != prev, vmg);
|
| 1053 |
|
| 1054 | if (vmg->__remove_middle) {
|
| 1055 | vmg->end = next->vm_end;
|
| 1056 | vmg->pgoff = vma_start_pgoff(next) - pglen;
|
| 1057 | vmg->anon_pgoff = vma_start_anon_pgoff(next) - pglen;
|
| 1058 | } else {
|
| 1059 |
|
| 1060 | vmg->__adjust_next_start = true;
|
| 1061 | vmg->start = middle->vm_start;
|
| 1062 | vmg->end = start;
|
| 1063 | vmg->pgoff = vma_start_pgoff(middle);
|
| 1064 | vmg->anon_pgoff = vma_start_anon_pgoff(middle);
|
| 1065 | }
|
| 1066 |
|
| 1067 | err = dup_anon_vma(next, middle, &anon_dup);
|
| 1068 | }
|
| 1069 |
|
| 1070 | if (err || commit_merge(vmg))
|
| 1071 | goto abort;
|
| 1072 |
|
| 1073 | vma_set_flags_mask(vmg->target, sticky_flags);
|
| 1074 | khugepaged_enter_vma(vmg->target, vmg->vm_flags);
|
| 1075 | vmg->state = VMA_MERGE_SUCCESS;
|
| 1076 | return vmg->target;
|
| 1077 |
|
| 1078 | abort:
|
| 1079 | vma_iter_set(vmg->vmi, start);
|
| 1080 | vma_iter_load(vmg->vmi);
|
| 1081 |
|
| 1082 | if (anon_dup)
|
| 1083 | unlink_anon_vmas(anon_dup);
|
| 1084 |
|
| 1085 | |
| 1086 | * This means we have failed to clone anon_vma's correctly, but no
|
| 1087 | * actual changes to VMAs have occurred, so no harm no foul - if the
|
| 1088 | * user doesn't want this reported and instead just wants to give up on
|
| 1089 | * the merge, allow it.
|
| 1090 |
|
| 1091 | if (!vmg->give_up_on_oom)
|
| 1092 | vmg->state = VMA_MERGE_ERROR_NOMEM;
|
| 1093 | return NULL;
|
| 1094 | }
|
| 1095 |
|
| 1096 | |
| 1097 | * vma_merge_new_range - Attempt to merge a new VMA into address space
|
| 1098 | *
|
| 1099 | * @vmg: Describes the VMA we are adding, in the range @vmg->start to @vmg->end
|
| 1100 | * (exclusive), which we try to merge with any adjacent VMAs if possible.
|
| 1101 | *
|
| 1102 | * We are about to add a VMA to the address space starting at @vmg->start and
|
| 1103 | * ending at @vmg->end. There are three different possible scenarios:
|
| 1104 | *
|
| 1105 | * 1. There is a VMA with identical properties immediately adjacent to the
|
| 1106 | * proposed new VMA [@vmg->start, @vmg->end) either before or after it -
|
| 1107 | * EXPAND that VMA:
|
| 1108 | *
|
| 1109 | * Proposed: |-----| or |-----|
|
| 1110 | * Existing: |----| |----|
|
| 1111 | *
|
| 1112 | * 2. There are VMAs with identical properties immediately adjacent to the
|
| 1113 | * proposed new VMA [@vmg->start, @vmg->end) both before AND after it -
|
| 1114 | * EXPAND the former and REMOVE the latter:
|
| 1115 | *
|
| 1116 | * Proposed: |-----|
|
| 1117 | * Existing: |----| |----|
|
| 1118 | *
|
| 1119 | * 3. There are no VMAs immediately adjacent to the proposed new VMA or those
|
| 1120 | * VMAs do not have identical attributes - NO MERGE POSSIBLE.
|
| 1121 | *
|
| 1122 | * In instances where we can merge, this function returns the expanded VMA which
|
| 1123 | * will have its range adjusted accordingly and the underlying maple tree also
|
| 1124 | * adjusted.
|
| 1125 | *
|
| 1126 | * Returns: In instances where no merge was possible, NULL. Otherwise, a pointer
|
| 1127 | * to the VMA we expanded.
|
| 1128 | *
|
| 1129 | * This function adjusts @vmg to provide @vmg->next if not already specified,
|
| 1130 | * and adjusts [@vmg->start, @vmg->end) to span the expanded range.
|
| 1131 | *
|
| 1132 | * ASSUMPTIONS:
|
| 1133 | * - The caller must hold a WRITE lock on the mm_struct->mmap_lock.
|
| 1134 | * - The caller must have determined that [@vmg->start, @vmg->end) is empty,
|
| 1135 | other than VMAs that will be unmapped should the operation succeed.
|
| 1136 | * - The caller must have specified the previous vma in @vmg->prev.
|
| 1137 | * - The caller must have specified the next vma in @vmg->next.
|
| 1138 | * - The caller must have positioned the vmi at or before the gap.
|
| 1139 |
|
| 1140 | struct vm_area_struct *vma_merge_new_range(struct vma_merge_struct *vmg)
|
| 1141 | {
|
| 1142 | struct vm_area_struct *prev = vmg->prev;
|
| 1143 | struct vm_area_struct *next = vmg->next;
|
| 1144 | unsigned long end = vmg->end;
|
| 1145 | bool can_merge_left, can_merge_right;
|
| 1146 |
|
| 1147 | mmap_assert_write_locked(vmg->mm);
|
| 1148 | VM_WARN_ON_VMG(vmg->middle, vmg);
|
| 1149 | VM_WARN_ON_VMG(vmg->target, vmg);
|
| 1150 |
|
| 1151 | VM_WARN_ON_VMG(vma_iter_addr(vmg->vmi) > end, vmg);
|
| 1152 |
|
| 1153 | vmg->state = VMA_MERGE_NOMERGE;
|
| 1154 |
|
| 1155 |
|
| 1156 | if (vma_flags_test_any_mask(&vmg->vma_flags, VMA_SPECIAL_FLAGS) ||
|
| 1157 | (!prev && !next))
|
| 1158 | return NULL;
|
| 1159 |
|
| 1160 | can_merge_left = can_vma_merge_left(vmg);
|
| 1161 | can_merge_right = !vmg->just_expand && can_vma_merge_right(vmg, can_merge_left);
|
| 1162 |
|
| 1163 |
|
| 1164 | if (can_merge_right) {
|
| 1165 | vmg->end = next->vm_end;
|
| 1166 | vmg->target = next;
|
| 1167 | }
|
| 1168 |
|
| 1169 |
|
| 1170 | if (can_merge_left) {
|
| 1171 | vmg->start = prev->vm_start;
|
| 1172 | vmg->target = prev;
|
| 1173 | vmg->pgoff = vma_start_pgoff(prev);
|
| 1174 | vmg->anon_pgoff = vma_start_anon_pgoff(prev);
|
| 1175 |
|
| 1176 | |
| 1177 | * If this merge would result in removal of the next VMA but we
|
| 1178 | * are not permitted to do so, reduce the operation to merging
|
| 1179 | * prev and vma.
|
| 1180 |
|
| 1181 | if (can_merge_right && !can_merge_remove_vma(next))
|
| 1182 | vmg->end = end;
|
| 1183 |
|
| 1184 |
|
| 1185 | if (!vmg->just_expand) {
|
| 1186 |
|
| 1187 | vma_prev(vmg->vmi);
|
| 1188 | }
|
| 1189 | }
|
| 1190 |
|
| 1191 | |
| 1192 | * Now try to expand adjacent VMA(s). This takes care of removing the
|
| 1193 | * following VMA if we have VMAs on both sides.
|
| 1194 |
|
| 1195 | if (vmg->target && !vma_expand(vmg)) {
|
| 1196 | khugepaged_enter_vma(vmg->target, vmg->vm_flags);
|
| 1197 | vmg->state = VMA_MERGE_SUCCESS;
|
| 1198 | return vmg->target;
|
| 1199 | }
|
| 1200 |
|
| 1201 | return NULL;
|
| 1202 | }
|
| 1203 |
|
| 1204 | |
| 1205 | * vma_merge_copied_range - Attempt to merge a VMA that is being copied by
|
| 1206 | * mremap()
|
| 1207 | *
|
| 1208 | * @vmg: Describes the VMA we are adding, in the copied-to range @vmg->start to
|
| 1209 | * @vmg->end (exclusive), which we try to merge with any adjacent VMAs if
|
| 1210 | * possible.
|
| 1211 | *
|
| 1212 | * vmg->prev, next, start, end, pgoff should all be relative to the COPIED TO
|
| 1213 | * range, i.e. the target range for the VMA.
|
| 1214 | *
|
| 1215 | * Returns: In instances where no merge was possible, NULL. Otherwise, a pointer
|
| 1216 | * to the VMA we expanded.
|
| 1217 | *
|
| 1218 | * ASSUMPTIONS: Same as vma_merge_new_range(), except vmg->middle must contain
|
| 1219 | * the copied-from VMA.
|
| 1220 |
|
| 1221 | static struct vm_area_struct *vma_merge_copied_range(struct vma_merge_struct *vmg)
|
| 1222 | {
|
| 1223 |
|
| 1224 | VM_WARN_ON_VMG(!vmg->middle, vmg);
|
| 1225 |
|
| 1226 | vmg->copied_from = vmg->middle;
|
| 1227 | vmg->middle = NULL;
|
| 1228 | return vma_merge_new_range(vmg);
|
| 1229 | }
|
| 1230 |
|
| 1231 | |
| 1232 | * vma_expand - Expand an existing VMA
|
| 1233 | *
|
| 1234 | * @vmg: Describes a VMA expansion operation.
|
| 1235 | *
|
| 1236 | * Expand @vma to vmg->start and vmg->end. Can expand off the start and end.
|
| 1237 | * Will expand over vmg->next if it's different from vmg->target and vmg->end ==
|
| 1238 | * vmg->next->vm_end. Checking if the vmg->target can expand and merge with
|
| 1239 | * vmg->next needs to be handled by the caller.
|
| 1240 | *
|
| 1241 | * Returns: 0 on success.
|
| 1242 | *
|
| 1243 | * ASSUMPTIONS:
|
| 1244 | * - The caller must hold a WRITE lock on the mm_struct->mmap_lock.
|
| 1245 | * - The caller must have set @vmg->target and @vmg->next.
|
| 1246 |
|
| 1247 | int vma_expand(struct vma_merge_struct *vmg)
|
| 1248 | {
|
| 1249 | struct vm_area_struct *anon_dup = NULL;
|
| 1250 | struct vm_area_struct *target = vmg->target;
|
| 1251 | struct vm_area_struct *next = vmg->next;
|
| 1252 | bool remove_next = false;
|
| 1253 | vma_flags_t sticky_flags =
|
| 1254 | vma_flags_and_mask(&vmg->vma_flags, VMA_STICKY_FLAGS);
|
| 1255 | vma_flags_t target_sticky;
|
| 1256 | int ret = 0;
|
| 1257 |
|
| 1258 | mmap_assert_write_locked(vmg->mm);
|
| 1259 | vma_start_write(target);
|
| 1260 |
|
| 1261 | target_sticky = vma_flags_and_mask(&target->flags, VMA_STICKY_FLAGS);
|
| 1262 |
|
| 1263 | if (next && target != next && vmg->end == next->vm_end)
|
| 1264 | remove_next = true;
|
| 1265 |
|
| 1266 |
|
| 1267 | VM_WARN_ON_VMG(!target, vmg);
|
| 1268 |
|
| 1269 | VM_WARN_ON_VMG(remove_next && !can_merge_remove_vma(next), vmg);
|
| 1270 |
|
| 1271 | VM_WARN_ON_VMG(next && !remove_next &&
|
| 1272 | next != target && vmg->end > next->vm_start, vmg);
|
| 1273 |
|
| 1274 | VM_WARN_ON_VMG(target->vm_start < vmg->start ||
|
| 1275 | target->vm_end > vmg->end, vmg);
|
| 1276 |
|
| 1277 | vma_flags_set_mask(&sticky_flags, target_sticky);
|
| 1278 |
|
| 1279 | |
| 1280 | * If we are removing the next VMA or copying from a VMA
|
| 1281 | * (e.g. mremap()'ing), we must propagate anon_vma state.
|
| 1282 | *
|
| 1283 | * Note that, by convention, callers ignore OOM for this case, so
|
| 1284 | * we don't need to account for vmg->give_up_on_mm here.
|
| 1285 |
|
| 1286 | if (remove_next)
|
| 1287 | ret = dup_anon_vma(target, next, &anon_dup);
|
| 1288 | if (!ret && vmg->copied_from)
|
| 1289 | ret = dup_anon_vma(target, vmg->copied_from, &anon_dup);
|
| 1290 | if (ret)
|
| 1291 | return ret;
|
| 1292 |
|
| 1293 | if (remove_next) {
|
| 1294 | vma_flags_t next_sticky;
|
| 1295 |
|
| 1296 | vma_start_write(next);
|
| 1297 | vmg->__remove_next = true;
|
| 1298 |
|
| 1299 | next_sticky = vma_flags_and_mask(&next->flags, VMA_STICKY_FLAGS);
|
| 1300 | vma_flags_set_mask(&sticky_flags, next_sticky);
|
| 1301 | }
|
| 1302 | if (commit_merge(vmg))
|
| 1303 | goto nomem;
|
| 1304 |
|
| 1305 | vma_set_flags_mask(target, sticky_flags);
|
| 1306 | return 0;
|
| 1307 |
|
| 1308 | nomem:
|
| 1309 | if (anon_dup)
|
| 1310 | unlink_anon_vmas(anon_dup);
|
| 1311 | |
| 1312 | * If the user requests that we just give upon OOM, we are safe to do so
|
| 1313 | * here, as commit merge provides this contract to us. Nothing has been
|
| 1314 | * changed - no harm no foul, just don't report it.
|
| 1315 |
|
| 1316 | if (!vmg->give_up_on_oom)
|
| 1317 | vmg->state = VMA_MERGE_ERROR_NOMEM;
|
| 1318 | return -ENOMEM;
|
| 1319 | }
|
| 1320 |
|
| 1321 | |
| 1322 | * vma_shrink() - Shrink the end of a VMA
|
| 1323 | * @vmi: The vma iterator
|
| 1324 | * @vma: The VMA to modify
|
| 1325 | * @end: The new end
|
| 1326 | *
|
| 1327 | * Note that the caller may only shrink the end of the VMA.
|
| 1328 | *
|
| 1329 | * Returns: 0 on success, -ENOMEM otherwise
|
| 1330 |
|
| 1331 | int vma_shrink(struct vma_iterator *vmi, struct vm_area_struct *vma,
|
| 1332 | unsigned long end)
|
| 1333 | {
|
| 1334 | struct vma_prepare vp;
|
| 1335 |
|
| 1336 | VM_WARN_ON_ONCE(end > vma->vm_end);
|
| 1337 |
|
| 1338 | vma_iter_config(vmi, end, vma->vm_end);
|
| 1339 | if (vma_iter_prealloc(vmi, NULL))
|
| 1340 | return -ENOMEM;
|
| 1341 |
|
| 1342 | vma_start_write(vma);
|
| 1343 |
|
| 1344 | init_vma_prep(&vp, vma);
|
| 1345 | vma_prepare(&vp);
|
| 1346 | vma_adjust_trans_huge(vma, vma->vm_start, end, NULL);
|
| 1347 |
|
| 1348 | vma_iter_clear(vmi);
|
| 1349 | __vma_set_range(vma, vma->vm_start, end);
|
| 1350 | vma_complete(&vp, vmi, vma->vm_mm);
|
| 1351 | validate_mm(vma->vm_mm);
|
| 1352 | return 0;
|
| 1353 | }
|
| 1354 |
|
| 1355 | static inline void vms_clear_ptes(struct vma_munmap_struct *vms,
|
| 1356 | struct ma_state *mas_detach, bool mm_wr_locked)
|
| 1357 | {
|
| 1358 | struct unmap_desc unmap = {
|
| 1359 | .mas = mas_detach,
|
| 1360 | .first = vms->vma,
|
| 1361 |
|
| 1362 | .pg_start = vms->unmap_start,
|
| 1363 | .pg_end = vms->unmap_end,
|
| 1364 | .vma_start = vms->start,
|
| 1365 | .vma_end = vms->end,
|
| 1366 | |
| 1367 | * The tree limits and reset differ from the normal case since it's a
|
| 1368 | * side-tree
|
| 1369 |
|
| 1370 | .tree_reset = 1,
|
| 1371 | .tree_end = vms->vma_count,
|
| 1372 | |
| 1373 | * We can free page tables without write-locking mmap_lock because VMAs
|
| 1374 | * were isolated before we downgraded mmap_lock.
|
| 1375 |
|
| 1376 | .mm_wr_locked = mm_wr_locked,
|
| 1377 | };
|
| 1378 |
|
| 1379 | if (!vms->clear_ptes)
|
| 1380 | return;
|
| 1381 |
|
| 1382 | mas_set(mas_detach, 1);
|
| 1383 | unmap_region(&unmap);
|
| 1384 | vms->clear_ptes = false;
|
| 1385 | }
|
| 1386 |
|
| 1387 | static void vms_clean_up_area(struct vma_munmap_struct *vms,
|
| 1388 | struct ma_state *mas_detach)
|
| 1389 | {
|
| 1390 | struct vm_area_struct *vma;
|
| 1391 |
|
| 1392 | if (!vms->nr_pages)
|
| 1393 | return;
|
| 1394 |
|
| 1395 | vms_clear_ptes(vms, mas_detach, true);
|
| 1396 | mas_set(mas_detach, 0);
|
| 1397 | mas_for_each(mas_detach, vma, ULONG_MAX)
|
| 1398 | vma_close(vma);
|
| 1399 | }
|
| 1400 |
|
| 1401 | |
| 1402 | * vms_complete_munmap_vmas() - Finish the munmap() operation
|
| 1403 | * @vms: The vma munmap struct
|
| 1404 | * @mas_detach: The maple state of the detached vmas
|
| 1405 | *
|
| 1406 | * This updates the mm_struct, unmaps the region, frees the resources
|
| 1407 | * used for the munmap() and may downgrade the lock - if requested. Everything
|
| 1408 | * needed to be done once the vma maple tree is updated.
|
| 1409 |
|
| 1410 | static void vms_complete_munmap_vmas(struct vma_munmap_struct *vms,
|
| 1411 | struct ma_state *mas_detach)
|
| 1412 | {
|
| 1413 | struct vm_area_struct *vma;
|
| 1414 | struct mm_struct *mm;
|
| 1415 |
|
| 1416 | mm = current->mm;
|
| 1417 | mm->map_count -= vms->vma_count;
|
| 1418 | mm->locked_vm -= vms->locked_vm;
|
| 1419 | if (vms->unlock)
|
| 1420 | mmap_write_downgrade(mm);
|
| 1421 |
|
| 1422 | if (!vms->nr_pages)
|
| 1423 | return;
|
| 1424 |
|
| 1425 | vms_clear_ptes(vms, mas_detach, !vms->unlock);
|
| 1426 |
|
| 1427 | update_hiwater_vm(mm);
|
| 1428 |
|
| 1429 | WRITE_ONCE(mm->total_vm, READ_ONCE(mm->total_vm) - vms->nr_pages);
|
| 1430 |
|
| 1431 | VM_WARN_ON(vms->exec_vm > mm->exec_vm);
|
| 1432 | VM_WARN_ON(vms->stack_vm > mm->stack_vm);
|
| 1433 | VM_WARN_ON(vms->data_vm > mm->data_vm);
|
| 1434 | mm->exec_vm -= vms->exec_vm;
|
| 1435 | mm->stack_vm -= vms->stack_vm;
|
| 1436 | mm->data_vm -= vms->data_vm;
|
| 1437 |
|
| 1438 |
|
| 1439 | mas_set(mas_detach, 0);
|
| 1440 | mas_for_each(mas_detach, vma, ULONG_MAX)
|
| 1441 | remove_vma(vma);
|
| 1442 |
|
| 1443 | vm_unacct_memory(vms->nr_accounted);
|
| 1444 | validate_mm(mm);
|
| 1445 | if (vms->unlock)
|
| 1446 | mmap_read_unlock(mm);
|
| 1447 |
|
| 1448 | __mt_destroy(mas_detach->tree);
|
| 1449 | }
|
| 1450 |
|
| 1451 | |
| 1452 | * reattach_vmas() - Undo any munmap work and free resources
|
| 1453 | * @mas_detach: The maple state with the detached maple tree
|
| 1454 | *
|
| 1455 | * Reattach any detached vmas and free up the maple tree used to track the vmas.
|
| 1456 |
|
| 1457 | static void reattach_vmas(struct ma_state *mas_detach)
|
| 1458 | {
|
| 1459 | struct vm_area_struct *vma;
|
| 1460 |
|
| 1461 | mas_set(mas_detach, 0);
|
| 1462 | mas_for_each(mas_detach, vma, ULONG_MAX)
|
| 1463 | vma_mark_attached(vma);
|
| 1464 |
|
| 1465 | __mt_destroy(mas_detach->tree);
|
| 1466 | }
|
| 1467 |
|
| 1468 | |
| 1469 | * vms_gather_munmap_vmas() - Put all VMAs within a range into a maple tree
|
| 1470 | * for removal at a later date. Handles splitting first and last if necessary
|
| 1471 | * and marking the vmas as isolated.
|
| 1472 | *
|
| 1473 | * @vms: The vma munmap struct
|
| 1474 | * @mas_detach: The maple state tracking the detached tree
|
| 1475 | *
|
| 1476 | * Return: 0 on success, error otherwise
|
| 1477 |
|
| 1478 | static int vms_gather_munmap_vmas(struct vma_munmap_struct *vms,
|
| 1479 | struct ma_state *mas_detach)
|
| 1480 | {
|
| 1481 | struct vm_area_struct *next = NULL;
|
| 1482 | int error;
|
| 1483 |
|
| 1484 | |
| 1485 | * If we need to split any vma, do it now to save pain later.
|
| 1486 | * Does it split the first one?
|
| 1487 |
|
| 1488 | if (vms->start > vms->vma->vm_start) {
|
| 1489 |
|
| 1490 | |
| 1491 | * Make sure that map_count on return from munmap() will
|
| 1492 | * not exceed its limit; but let map_count go just above
|
| 1493 | * its limit temporarily, to help free resources as expected.
|
| 1494 |
|
| 1495 | if (vms->end < vms->vma->vm_end &&
|
| 1496 | vms->vma->vm_mm->map_count >= get_sysctl_max_map_count()) {
|
| 1497 | error = -ENOMEM;
|
| 1498 | goto map_count_exceeded;
|
| 1499 | }
|
| 1500 |
|
| 1501 |
|
| 1502 | if (vma_is_sealed(vms->vma)) {
|
| 1503 | error = -EPERM;
|
| 1504 | goto start_split_failed;
|
| 1505 | }
|
| 1506 |
|
| 1507 | error = __split_vma(vms->vmi, vms->vma, vms->start, 1);
|
| 1508 | if (error)
|
| 1509 | goto start_split_failed;
|
| 1510 | }
|
| 1511 | vms->prev = vma_prev(vms->vmi);
|
| 1512 | if (vms->prev)
|
| 1513 | vms->unmap_start = vms->prev->vm_end;
|
| 1514 |
|
| 1515 | |
| 1516 | * Detach a range of VMAs from the mm. Using next as a temp variable as
|
| 1517 | * it is always overwritten.
|
| 1518 |
|
| 1519 | for_each_vma_range(*(vms->vmi), next, vms->end) {
|
| 1520 | long nrpages;
|
| 1521 |
|
| 1522 | if (vma_is_sealed(next)) {
|
| 1523 | error = -EPERM;
|
| 1524 | goto modify_vma_failed;
|
| 1525 | }
|
| 1526 |
|
| 1527 | if (next->vm_end > vms->end) {
|
| 1528 | error = __split_vma(vms->vmi, next, vms->end, 0);
|
| 1529 | if (error)
|
| 1530 | goto end_split_failed;
|
| 1531 | }
|
| 1532 | vma_start_write(next);
|
| 1533 | mas_set(mas_detach, vms->vma_count++);
|
| 1534 | error = mas_store_gfp(mas_detach, next, GFP_KERNEL);
|
| 1535 | if (error)
|
| 1536 | goto munmap_gather_failed;
|
| 1537 |
|
| 1538 | vma_mark_detached(next);
|
| 1539 | nrpages = vma_pages(next);
|
| 1540 |
|
| 1541 | vms->nr_pages += nrpages;
|
| 1542 | if (vma_test(next, VMA_LOCKED_BIT))
|
| 1543 | vms->locked_vm += nrpages;
|
| 1544 |
|
| 1545 | if (vma_test(next, VMA_ACCOUNT_BIT))
|
| 1546 | vms->nr_accounted += nrpages;
|
| 1547 |
|
| 1548 | if (is_exec_mapping(next->vm_flags))
|
| 1549 | vms->exec_vm += nrpages;
|
| 1550 | else if (is_stack_mapping(next->vm_flags))
|
| 1551 | vms->stack_vm += nrpages;
|
| 1552 | else if (is_data_mapping_vma_flags(&next->flags))
|
| 1553 | vms->data_vm += nrpages;
|
| 1554 |
|
| 1555 | if (vms->uf) {
|
| 1556 | |
| 1557 | * If userfaultfd_unmap_prep returns an error the vmas
|
| 1558 | * will remain split, but userland will get a
|
| 1559 | * highly unexpected error anyway. This is no
|
| 1560 | * different than the case where the first of the two
|
| 1561 | * __split_vma fails, but we don't undo the first
|
| 1562 | * split, despite we could. This is unlikely enough
|
| 1563 | * failure that it's not worth optimizing it for.
|
| 1564 |
|
| 1565 | error = userfaultfd_unmap_prep(next, vms->start,
|
| 1566 | vms->end, vms->uf);
|
| 1567 | if (error)
|
| 1568 | goto userfaultfd_error;
|
| 1569 | }
|
| 1570 | #ifdef CONFIG_DEBUG_VM_MAPLE_TREE
|
| 1571 | BUG_ON(next->vm_start < vms->start);
|
| 1572 | BUG_ON(next->vm_start > vms->end);
|
| 1573 | #endif
|
| 1574 | }
|
| 1575 |
|
| 1576 | vms->next = vma_next(vms->vmi);
|
| 1577 | if (vms->next)
|
| 1578 | vms->unmap_end = vms->next->vm_start;
|
| 1579 |
|
| 1580 | #if defined(CONFIG_DEBUG_VM_MAPLE_TREE)
|
| 1581 |
|
| 1582 | {
|
| 1583 | MA_STATE(test, mas_detach->tree, 0, 0);
|
| 1584 | struct vm_area_struct *vma_mas, *vma_test;
|
| 1585 | int test_count = 0;
|
| 1586 |
|
| 1587 | vma_iter_set(vms->vmi, vms->start);
|
| 1588 | rcu_read_lock();
|
| 1589 | vma_test = mas_find(&test, vms->vma_count - 1);
|
| 1590 | for_each_vma_range(*(vms->vmi), vma_mas, vms->end) {
|
| 1591 | BUG_ON(vma_mas != vma_test);
|
| 1592 | test_count++;
|
| 1593 | vma_test = mas_next(&test, vms->vma_count - 1);
|
| 1594 | }
|
| 1595 | rcu_read_unlock();
|
| 1596 | BUG_ON(vms->vma_count != test_count);
|
| 1597 | }
|
| 1598 | #endif
|
| 1599 |
|
| 1600 | while (vma_iter_addr(vms->vmi) > vms->start)
|
| 1601 | vma_iter_prev_range(vms->vmi);
|
| 1602 |
|
| 1603 | vms->clear_ptes = true;
|
| 1604 | return 0;
|
| 1605 |
|
| 1606 | userfaultfd_error:
|
| 1607 | munmap_gather_failed:
|
| 1608 | end_split_failed:
|
| 1609 | modify_vma_failed:
|
| 1610 | reattach_vmas(mas_detach);
|
| 1611 | start_split_failed:
|
| 1612 | map_count_exceeded:
|
| 1613 | return error;
|
| 1614 | }
|
| 1615 |
|
| 1616 | |
| 1617 | * init_vma_munmap() - Initializer wrapper for vma_munmap_struct
|
| 1618 | * @vms: The vma munmap struct
|
| 1619 | * @vmi: The vma iterator
|
| 1620 | * @vma: The first vm_area_struct to munmap
|
| 1621 | * @start: The aligned start address to munmap
|
| 1622 | * @end: The aligned end address to munmap
|
| 1623 | * @uf: The userfaultfd list_head
|
| 1624 | * @unlock: Unlock after the operation. Only unlocked on success
|
| 1625 |
|
| 1626 | static void init_vma_munmap(struct vma_munmap_struct *vms,
|
| 1627 | struct vma_iterator *vmi, struct vm_area_struct *vma,
|
| 1628 | unsigned long start, unsigned long end, struct list_head *uf,
|
| 1629 | bool unlock)
|
| 1630 | {
|
| 1631 | vms->vmi = vmi;
|
| 1632 | vms->vma = vma;
|
| 1633 | if (vma) {
|
| 1634 | vms->start = start;
|
| 1635 | vms->end = end;
|
| 1636 | } else {
|
| 1637 | vms->start = vms->end = 0;
|
| 1638 | }
|
| 1639 | vms->unlock = unlock;
|
| 1640 | vms->uf = uf;
|
| 1641 | vms->vma_count = 0;
|
| 1642 | vms->nr_pages = vms->locked_vm = vms->nr_accounted = 0;
|
| 1643 | vms->exec_vm = vms->stack_vm = vms->data_vm = 0;
|
| 1644 | vms->unmap_start = FIRST_USER_ADDRESS;
|
| 1645 | vms->unmap_end = USER_PGTABLES_CEILING;
|
| 1646 | vms->clear_ptes = false;
|
| 1647 | }
|
| 1648 |
|
| 1649 | |
| 1650 | * do_vmi_align_munmap() - munmap the aligned region from @start to @end.
|
| 1651 | * @vmi: The vma iterator
|
| 1652 | * @vma: The starting vm_area_struct
|
| 1653 | * @mm: The mm_struct
|
| 1654 | * @start: The aligned start address to munmap.
|
| 1655 | * @end: The aligned end address to munmap.
|
| 1656 | * @uf: The userfaultfd list_head
|
| 1657 | * @unlock: Set to true to drop the mmap_lock. unlocking only happens on
|
| 1658 | * success.
|
| 1659 | *
|
| 1660 | * Return: 0 on success and drops the lock if so directed, error and leaves the
|
| 1661 | * lock held otherwise.
|
| 1662 |
|
| 1663 | int do_vmi_align_munmap(struct vma_iterator *vmi, struct vm_area_struct *vma,
|
| 1664 | struct mm_struct *mm, unsigned long start, unsigned long end,
|
| 1665 | struct list_head *uf, bool unlock)
|
| 1666 | {
|
| 1667 | struct maple_tree mt_detach;
|
| 1668 | MA_STATE(mas_detach, &mt_detach, 0, 0);
|
| 1669 | mt_init_flags(&mt_detach, vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK);
|
| 1670 | mt_on_stack(mt_detach);
|
| 1671 | struct vma_munmap_struct vms;
|
| 1672 | int error;
|
| 1673 |
|
| 1674 | init_vma_munmap(&vms, vmi, vma, start, end, uf, unlock);
|
| 1675 | error = vms_gather_munmap_vmas(&vms, &mas_detach);
|
| 1676 | if (error)
|
| 1677 | goto gather_failed;
|
| 1678 |
|
| 1679 | error = vma_iter_clear_gfp(vmi, start, end, GFP_KERNEL);
|
| 1680 | if (error)
|
| 1681 | goto clear_tree_failed;
|
| 1682 |
|
| 1683 |
|
| 1684 | vms_complete_munmap_vmas(&vms, &mas_detach);
|
| 1685 | return 0;
|
| 1686 |
|
| 1687 | clear_tree_failed:
|
| 1688 | reattach_vmas(&mas_detach);
|
| 1689 | gather_failed:
|
| 1690 | validate_mm(mm);
|
| 1691 | return error;
|
| 1692 | }
|
| 1693 |
|
| 1694 | |
| 1695 | * do_vmi_munmap() - munmap a given range.
|
| 1696 | * @vmi: The vma iterator
|
| 1697 | * @mm: The mm_struct
|
| 1698 | * @start: The start address to munmap
|
| 1699 | * @len: The length of the range to munmap
|
| 1700 | * @uf: The userfaultfd list_head
|
| 1701 | * @unlock: set to true if the user wants to drop the mmap_lock on success
|
| 1702 | *
|
| 1703 | * This function takes a @mas that is either pointing to the previous VMA or set
|
| 1704 | * to MA_START and sets it up to remove the mapping(s). The @len will be
|
| 1705 | * aligned.
|
| 1706 | *
|
| 1707 | * Return: 0 on success and drops the lock if so directed, error and leaves the
|
| 1708 | * lock held otherwise.
|
| 1709 |
|
| 1710 | int do_vmi_munmap(struct vma_iterator *vmi, struct mm_struct *mm,
|
| 1711 | unsigned long start, size_t len, struct list_head *uf,
|
| 1712 | bool unlock)
|
| 1713 | {
|
| 1714 | unsigned long end;
|
| 1715 | struct vm_area_struct *vma;
|
| 1716 |
|
| 1717 | if ((offset_in_page(start)) || start > TASK_SIZE || len > TASK_SIZE-start)
|
| 1718 | return -EINVAL;
|
| 1719 |
|
| 1720 | end = start + PAGE_ALIGN(len);
|
| 1721 | if (end == start)
|
| 1722 | return -EINVAL;
|
| 1723 |
|
| 1724 |
|
| 1725 | vma = vma_find(vmi, end);
|
| 1726 | if (!vma) {
|
| 1727 | if (unlock)
|
| 1728 | mmap_write_unlock(mm);
|
| 1729 | return 0;
|
| 1730 | }
|
| 1731 |
|
| 1732 | return do_vmi_align_munmap(vmi, vma, mm, start, end, uf, unlock);
|
| 1733 | }
|
| 1734 |
|
| 1735 | |
| 1736 | * We are about to modify one or multiple of a VMA's flags, policy, userfaultfd
|
| 1737 | * context and anonymous VMA name within the range [start, end).
|
| 1738 | *
|
| 1739 | * As a result, we might be able to merge the newly modified VMA range with an
|
| 1740 | * adjacent VMA with identical properties.
|
| 1741 | *
|
| 1742 | * If no merge is possible and the range does not span the entirety of the VMA,
|
| 1743 | * we then need to split the VMA to accommodate the change.
|
| 1744 | *
|
| 1745 | * The function returns either the merged VMA, the original VMA if a split was
|
| 1746 | * required instead, or an error if the split failed.
|
| 1747 |
|
| 1748 | static struct vm_area_struct *vma_modify(struct vma_merge_struct *vmg)
|
| 1749 | {
|
| 1750 | struct vm_area_struct *vma = vmg->middle;
|
| 1751 | unsigned long start = vmg->start;
|
| 1752 | unsigned long end = vmg->end;
|
| 1753 | struct vm_area_struct *merged;
|
| 1754 |
|
| 1755 |
|
| 1756 | merged = vma_merge_existing_range(vmg);
|
| 1757 | if (merged)
|
| 1758 | return merged;
|
| 1759 | if (vmg_nomem(vmg))
|
| 1760 | return ERR_PTR(-ENOMEM);
|
| 1761 |
|
| 1762 | |
| 1763 | * Split can fail for reasons other than OOM, so if the user requests
|
| 1764 | * this it's probably a mistake.
|
| 1765 |
|
| 1766 | VM_WARN_ON(vmg->give_up_on_oom &&
|
| 1767 | (vma->vm_start != start || vma->vm_end != end));
|
| 1768 |
|
| 1769 |
|
| 1770 | if (vma->vm_start < start) {
|
| 1771 | int err = split_vma(vmg->vmi, vma, start, 1);
|
| 1772 |
|
| 1773 | if (err)
|
| 1774 | return ERR_PTR(err);
|
| 1775 | }
|
| 1776 |
|
| 1777 |
|
| 1778 | if (vma->vm_end > end) {
|
| 1779 | int err = split_vma(vmg->vmi, vma, end, 0);
|
| 1780 |
|
| 1781 | if (err)
|
| 1782 | return ERR_PTR(err);
|
| 1783 | }
|
| 1784 |
|
| 1785 | return vma;
|
| 1786 | }
|
| 1787 |
|
| 1788 | struct vm_area_struct *vma_modify_flags(struct vma_iterator *vmi,
|
| 1789 | struct vm_area_struct *prev, struct vm_area_struct *vma,
|
| 1790 | unsigned long start, unsigned long end,
|
| 1791 | vma_flags_t *vma_flags_ptr)
|
| 1792 | {
|
| 1793 | VMG_VMA_STATE(vmg, vmi, prev, vma, start, end);
|
| 1794 | const vma_flags_t vma_flags = *vma_flags_ptr;
|
| 1795 | struct vm_area_struct *ret;
|
| 1796 |
|
| 1797 | vmg.vma_flags = vma_flags;
|
| 1798 |
|
| 1799 | ret = vma_modify(&vmg);
|
| 1800 | if (IS_ERR(ret))
|
| 1801 | return ret;
|
| 1802 |
|
| 1803 | |
| 1804 | * For a merge to succeed, the flags must match those
|
| 1805 | * requested. However, sticky flags may have been retained, so propagate
|
| 1806 | * them to the caller.
|
| 1807 |
|
| 1808 | if (vmg.state == VMA_MERGE_SUCCESS)
|
| 1809 | *vma_flags_ptr = ret->flags;
|
| 1810 | return ret;
|
| 1811 | }
|
| 1812 |
|
| 1813 | struct vm_area_struct *vma_modify_name(struct vma_iterator *vmi,
|
| 1814 | struct vm_area_struct *prev, struct vm_area_struct *vma,
|
| 1815 | unsigned long start, unsigned long end,
|
| 1816 | struct anon_vma_name *new_name)
|
| 1817 | {
|
| 1818 | VMG_VMA_STATE(vmg, vmi, prev, vma, start, end);
|
| 1819 |
|
| 1820 | vmg.anon_name = new_name;
|
| 1821 |
|
| 1822 | return vma_modify(&vmg);
|
| 1823 | }
|
| 1824 |
|
| 1825 | struct vm_area_struct *vma_modify_policy(struct vma_iterator *vmi,
|
| 1826 | struct vm_area_struct *prev, struct vm_area_struct *vma,
|
| 1827 | unsigned long start, unsigned long end,
|
| 1828 | struct mempolicy *new_pol)
|
| 1829 | {
|
| 1830 | VMG_VMA_STATE(vmg, vmi, prev, vma, start, end);
|
| 1831 |
|
| 1832 | vmg.policy = new_pol;
|
| 1833 |
|
| 1834 | return vma_modify(&vmg);
|
| 1835 | }
|
| 1836 |
|
| 1837 | struct vm_area_struct *vma_modify_flags_uffd(struct vma_iterator *vmi,
|
| 1838 | struct vm_area_struct *prev, struct vm_area_struct *vma,
|
| 1839 | unsigned long start, unsigned long end,
|
| 1840 | const vma_flags_t *vma_flags, struct vm_userfaultfd_ctx new_ctx,
|
| 1841 | bool give_up_on_oom)
|
| 1842 | {
|
| 1843 | VMG_VMA_STATE(vmg, vmi, prev, vma, start, end);
|
| 1844 |
|
| 1845 | vmg.vma_flags = *vma_flags;
|
| 1846 | vmg.uffd_ctx = new_ctx;
|
| 1847 | if (give_up_on_oom)
|
| 1848 | vmg.give_up_on_oom = true;
|
| 1849 |
|
| 1850 | return vma_modify(&vmg);
|
| 1851 | }
|
| 1852 |
|
| 1853 | |
| 1854 | * Expand vma by delta bytes, potentially merging with an immediately adjacent
|
| 1855 | * VMA with identical properties.
|
| 1856 |
|
| 1857 | struct vm_area_struct *vma_merge_extend(struct vma_iterator *vmi,
|
| 1858 | struct vm_area_struct *vma,
|
| 1859 | unsigned long delta)
|
| 1860 | {
|
| 1861 | VMG_VMA_STATE(vmg, vmi, vma, vma, vma->vm_end, vma->vm_end + delta);
|
| 1862 |
|
| 1863 | vmg.next = vma_iter_next_rewind(vmi, NULL);
|
| 1864 | vmg.middle = NULL;
|
| 1865 |
|
| 1866 | return vma_merge_new_range(&vmg);
|
| 1867 | }
|
| 1868 |
|
| 1869 | void unlink_file_vma_batch_init(struct unlink_vma_file_batch *vb)
|
| 1870 | {
|
| 1871 | vb->count = 0;
|
| 1872 | }
|
| 1873 |
|
| 1874 | static void unlink_file_vma_batch_process(struct unlink_vma_file_batch *vb)
|
| 1875 | {
|
| 1876 | struct address_space *mapping;
|
| 1877 | int i;
|
| 1878 |
|
| 1879 | mapping = vb->vmas[0]->vm_file->f_mapping;
|
| 1880 | i_mmap_lock_write(mapping);
|
| 1881 | for (i = 0; i < vb->count; i++) {
|
| 1882 | VM_WARN_ON_ONCE(vb->vmas[i]->vm_file->f_mapping != mapping);
|
| 1883 | __remove_shared_vm_struct(vb->vmas[i], mapping);
|
| 1884 | }
|
| 1885 | i_mmap_unlock_write(mapping);
|
| 1886 |
|
| 1887 | unlink_file_vma_batch_init(vb);
|
| 1888 | }
|
| 1889 |
|
| 1890 | void unlink_file_vma_batch_add(struct unlink_vma_file_batch *vb,
|
| 1891 | struct vm_area_struct *vma)
|
| 1892 | {
|
| 1893 | if (vma->vm_file == NULL)
|
| 1894 | return;
|
| 1895 |
|
| 1896 | if ((vb->count > 0 && vb->vmas[0]->vm_file != vma->vm_file) ||
|
| 1897 | vb->count == ARRAY_SIZE(vb->vmas))
|
| 1898 | unlink_file_vma_batch_process(vb);
|
| 1899 |
|
| 1900 | vb->vmas[vb->count] = vma;
|
| 1901 | vb->count++;
|
| 1902 | }
|
| 1903 |
|
| 1904 | void unlink_file_vma_batch_final(struct unlink_vma_file_batch *vb)
|
| 1905 | {
|
| 1906 | if (vb->count > 0)
|
| 1907 | unlink_file_vma_batch_process(vb);
|
| 1908 | }
|
| 1909 |
|
| 1910 | static void vma_link_file(struct vm_area_struct *vma, bool hold_rmap_lock)
|
| 1911 | {
|
| 1912 | struct file *file = vma->vm_file;
|
| 1913 | struct address_space *mapping;
|
| 1914 |
|
| 1915 | if (file) {
|
| 1916 | mapping = file->f_mapping;
|
| 1917 | i_mmap_lock_write(mapping);
|
| 1918 | __vma_link_file(vma, mapping);
|
| 1919 | if (!hold_rmap_lock)
|
| 1920 | i_mmap_unlock_write(mapping);
|
| 1921 | }
|
| 1922 | }
|
| 1923 |
|
| 1924 | static int vma_link(struct mm_struct *mm, struct vm_area_struct *vma)
|
| 1925 | {
|
| 1926 | VMA_ITERATOR(vmi, mm, 0);
|
| 1927 |
|
| 1928 | vma_iter_config(&vmi, vma->vm_start, vma->vm_end);
|
| 1929 | if (vma_iter_prealloc(&vmi, vma))
|
| 1930 | return -ENOMEM;
|
| 1931 |
|
| 1932 | vma_start_write(vma);
|
| 1933 | vma_iter_store_new(&vmi, vma);
|
| 1934 | vma_link_file(vma, false);
|
| 1935 | mm->map_count++;
|
| 1936 | validate_mm(mm);
|
| 1937 | return 0;
|
| 1938 | }
|
| 1939 |
|
| 1940 | |
| 1941 | * Copy the vma structure to a new location in the same mm,
|
| 1942 | * prior to moving page table entries, to effect an mremap move.
|
| 1943 |
|
| 1944 | struct vm_area_struct *copy_vma(struct vm_area_struct **vmap,
|
| 1945 | unsigned long addr, unsigned long len, pgoff_t pgoff,
|
| 1946 | pgoff_t anon_pgoff, bool *need_rmap_locks, bool keep_source)
|
| 1947 | {
|
| 1948 | struct vm_area_struct *vma = *vmap;
|
| 1949 | unsigned long old_vma_start = vma->vm_start;
|
| 1950 | struct mm_struct *mm = vma->vm_mm;
|
| 1951 | struct vm_area_struct *new_vma;
|
| 1952 | bool can_self_merge = false;
|
| 1953 | VMA_ITERATOR(vmi, mm, addr);
|
| 1954 | VMG_VMA_STATE(vmg, &vmi, NULL, vma, addr, addr + len);
|
| 1955 |
|
| 1956 | |
| 1957 | * If a vma has not yet been faulted, update its anonymous pgoff to
|
| 1958 | * match the new location to increase its chance of merging.
|
| 1959 |
|
| 1960 | if (!vma->anon_vma) {
|
| 1961 | anon_pgoff = addr >> PAGE_SHIFT;
|
| 1962 |
|
| 1963 | if (vma_is_anonymous(vma)) {
|
| 1964 | pgoff = anon_pgoff;
|
| 1965 | can_self_merge = true;
|
| 1966 | }
|
| 1967 | }
|
| 1968 |
|
| 1969 | |
| 1970 | * If the VMA we are copying might contain a uprobe PTE, ensure
|
| 1971 | * that we do not establish one upon merge. Otherwise, when mremap()
|
| 1972 | * moves page tables, it will orphan the newly created PTE.
|
| 1973 |
|
| 1974 | if (vma->vm_file)
|
| 1975 | vmg.skip_vma_uprobe = true;
|
| 1976 |
|
| 1977 | new_vma = find_vma_prev(mm, addr, &vmg.prev);
|
| 1978 | if (new_vma && new_vma->vm_start < addr + len)
|
| 1979 | return NULL;
|
| 1980 |
|
| 1981 | vmg.pgoff = pgoff;
|
| 1982 | vmg.anon_pgoff = anon_pgoff;
|
| 1983 | vmg.next = vma_iter_next_rewind(&vmi, NULL);
|
| 1984 |
|
| 1985 | |
| 1986 | * If the original VMA is kept (MREMAP_DONTUNMAP), the source and
|
| 1987 | * destination VMA must be treated distinctly.
|
| 1988 | *
|
| 1989 | * A merge violates this, so in this case disallow a self-merge.
|
| 1990 |
|
| 1991 | if (can_self_merge && keep_source) {
|
| 1992 | if (vmg.prev == vma)
|
| 1993 | vmg.prev = NULL;
|
| 1994 | if (vmg.next == vma)
|
| 1995 | vmg.next = NULL;
|
| 1996 | can_self_merge = false;
|
| 1997 | }
|
| 1998 |
|
| 1999 | new_vma = vma_merge_copied_range(&vmg);
|
| 2000 |
|
| 2001 | if (new_vma) {
|
| 2002 |
|
| 2003 | if (unlikely(new_vma->vm_start < old_vma_start &&
|
| 2004 | new_vma->vm_end > old_vma_start)) {
|
| 2005 | |
| 2006 | * The only way a VMA can both self-merge and be
|
| 2007 | * replaced is if the remap places the new VMA
|
| 2008 | * immediately prior to its old self ('next') and
|
| 2009 | * immediately after another VMA ('prev') causing the
|
| 2010 | * next to be removed and prev to be expanded to cover
|
| 2011 | * the entire range.
|
| 2012 | *
|
| 2013 | * This should only be possible if the anonymous page
|
| 2014 | * offset was updated, i.e. the VMA is unfaulted.
|
| 2015 |
|
| 2016 | VM_WARN_ON_ONCE_VMA(!can_self_merge, new_vma);
|
| 2017 | *vmap = vma = new_vma;
|
| 2018 | }
|
| 2019 | *need_rmap_locks =
|
| 2020 | (vma_start_pgoff(new_vma) <= vma_start_pgoff(vma));
|
| 2021 | } else {
|
| 2022 | new_vma = vm_area_dup(vma);
|
| 2023 | if (!new_vma)
|
| 2024 | goto out;
|
| 2025 | vma_set_range(new_vma, addr, addr + len, pgoff, anon_pgoff);
|
| 2026 | if (vma_dup_policy(vma, new_vma))
|
| 2027 | goto out_free_vma;
|
| 2028 | if (anon_vma_clone(new_vma, vma, VMA_OP_REMAP))
|
| 2029 | goto out_free_mempol;
|
| 2030 | if (new_vma->vm_file)
|
| 2031 | get_file(new_vma->vm_file);
|
| 2032 | if (new_vma->vm_ops && new_vma->vm_ops->open)
|
| 2033 | new_vma->vm_ops->open(new_vma);
|
| 2034 | if (vma_link(mm, new_vma))
|
| 2035 | goto out_vma_link;
|
| 2036 | *need_rmap_locks = false;
|
| 2037 | }
|
| 2038 | return new_vma;
|
| 2039 |
|
| 2040 | out_vma_link:
|
| 2041 | fixup_hugetlb_reservations(new_vma);
|
| 2042 | vma_close(new_vma);
|
| 2043 |
|
| 2044 | if (new_vma->vm_file)
|
| 2045 | fput(new_vma->vm_file);
|
| 2046 |
|
| 2047 | unlink_anon_vmas(new_vma);
|
| 2048 | out_free_mempol:
|
| 2049 | mpol_put(vma_policy(new_vma));
|
| 2050 | out_free_vma:
|
| 2051 | vm_area_free(new_vma);
|
| 2052 | out:
|
| 2053 | return NULL;
|
| 2054 | }
|
| 2055 |
|
| 2056 | |
| 2057 | * Rough compatibility check to quickly see if it's even worth looking
|
| 2058 | * at sharing an anon_vma.
|
| 2059 | *
|
| 2060 | * They need to have the same vm_file, and the flags can only differ
|
| 2061 | * in things that mprotect may change.
|
| 2062 | *
|
| 2063 | * NOTE! The fact that we share an anon_vma doesn't _have_ to mean that
|
| 2064 | * we can merge the two vma's. For example, we refuse to merge a vma if
|
| 2065 | * there is a vm_ops->close() function, because that indicates that the
|
| 2066 | * driver is doing some kind of reference counting. But that doesn't
|
| 2067 | * really matter for the anon_vma sharing case.
|
| 2068 |
|
| 2069 | static int anon_vma_compatible(struct vm_area_struct *a, struct vm_area_struct *b)
|
| 2070 | {
|
| 2071 | vma_flags_t diff = vma_flags_diff_pair(&a->flags, &b->flags);
|
| 2072 |
|
| 2073 |
|
| 2074 | vma_flags_clear_mask(&diff, VMA_ACCESS_FLAGS);
|
| 2075 |
|
| 2076 | vma_flags_clear_mask(&diff, VMA_IGNORE_MERGE_FLAGS);
|
| 2077 |
|
| 2078 |
|
| 2079 | if (a->vm_end != b->vm_start)
|
| 2080 | return false;
|
| 2081 |
|
| 2082 | if (!mpol_equal(vma_policy(a), vma_policy(b)))
|
| 2083 | return false;
|
| 2084 |
|
| 2085 | if (a->vm_file != b->vm_file)
|
| 2086 | return false;
|
| 2087 |
|
| 2088 | if (!vma_flags_empty(&diff))
|
| 2089 | return false;
|
| 2090 |
|
| 2091 | if (vma_end_pgoff(a) != vma_start_pgoff(b))
|
| 2092 | return false;
|
| 2093 |
|
| 2094 | if (vma_end_anon_pgoff(a) != vma_start_anon_pgoff(b))
|
| 2095 | return false;
|
| 2096 | return true;
|
| 2097 | }
|
| 2098 |
|
| 2099 | |
| 2100 | * Do some basic sanity checking to see if we can re-use the anon_vma
|
| 2101 | * from 'old'. The 'a'/'b' vma's are in VM order - one of them will be
|
| 2102 | * the same as 'old', the other will be the new one that is trying
|
| 2103 | * to share the anon_vma.
|
| 2104 | *
|
| 2105 | * NOTE! This runs with mmap_lock held for reading, so it is possible that
|
| 2106 | * the anon_vma of 'old' is concurrently in the process of being set up
|
| 2107 | * by another page fault trying to merge _that_. But that's ok: if it
|
| 2108 | * is being set up, that automatically means that it will be a singleton
|
| 2109 | * acceptable for merging, so we can do all of this optimistically. But
|
| 2110 | * we do that READ_ONCE() to make sure that we never re-load the pointer.
|
| 2111 | *
|
| 2112 | * The READ_ONCE() establishes an address dependency between anon_vma and
|
| 2113 | * any access to its fields, which pairs with the assignment to
|
| 2114 | * vma->anon_vma performed with release semantics in __anon_vma_prepare().
|
| 2115 | *
|
| 2116 | * This is especially important as anon_vma's are SLAB_TYPESAFE_BY_RCU so
|
| 2117 | * accessing an uninitialised anon_vma's fields may result in a UAF.
|
| 2118 | *
|
| 2119 | * IOW: that the "list_is_singular()" test on the anon_vma_chain only
|
| 2120 | * matters for the 'stable anon_vma' case (ie the thing we want to avoid
|
| 2121 | * is to return an anon_vma that is "complex" due to having gone through
|
| 2122 | * a fork).
|
| 2123 | *
|
| 2124 | * We also make sure that the two vma's are compatible (adjacent,
|
| 2125 | * and with the same memory policies). That's all stable, even with just
|
| 2126 | * a read lock on the mmap_lock.
|
| 2127 |
|
| 2128 | static struct anon_vma *reusable_anon_vma(struct vm_area_struct *old,
|
| 2129 | struct vm_area_struct *a,
|
| 2130 | struct vm_area_struct *b)
|
| 2131 | {
|
| 2132 | if (anon_vma_compatible(a, b)) {
|
| 2133 |
|
| 2134 | struct anon_vma *anon_vma = READ_ONCE(old->anon_vma);
|
| 2135 |
|
| 2136 | if (anon_vma && list_is_singular(&old->anon_vma_chain))
|
| 2137 | return anon_vma;
|
| 2138 | }
|
| 2139 | return NULL;
|
| 2140 | }
|
| 2141 |
|
| 2142 | |
| 2143 | * find_mergeable_anon_vma is used by anon_vma_prepare, to check
|
| 2144 | * neighbouring vmas for a suitable anon_vma, before it goes off
|
| 2145 | * to allocate a new anon_vma. It checks because a repetitive
|
| 2146 | * sequence of mprotects and faults may otherwise lead to distinct
|
| 2147 | * anon_vmas being allocated, preventing vma merge in subsequent
|
| 2148 | * mprotect.
|
| 2149 |
|
| 2150 | struct anon_vma *find_mergeable_anon_vma(struct vm_area_struct *vma)
|
| 2151 | {
|
| 2152 | struct anon_vma *anon_vma = NULL;
|
| 2153 | struct vm_area_struct *prev, *next;
|
| 2154 | VMA_ITERATOR(vmi, vma->vm_mm, vma->vm_end);
|
| 2155 |
|
| 2156 |
|
| 2157 | next = vma_iter_load(&vmi);
|
| 2158 | if (next) {
|
| 2159 | anon_vma = reusable_anon_vma(next, vma, next);
|
| 2160 | if (anon_vma)
|
| 2161 | return anon_vma;
|
| 2162 | }
|
| 2163 |
|
| 2164 | prev = vma_prev(&vmi);
|
| 2165 | VM_BUG_ON_VMA(prev != vma, vma);
|
| 2166 | prev = vma_prev(&vmi);
|
| 2167 |
|
| 2168 | if (prev)
|
| 2169 | anon_vma = reusable_anon_vma(prev, prev, vma);
|
| 2170 |
|
| 2171 | |
| 2172 | * We might reach here with anon_vma == NULL if we can't find
|
| 2173 | * any reusable anon_vma.
|
| 2174 | * There's no absolute need to look only at touching neighbours:
|
| 2175 | * we could search further afield for "compatible" anon_vmas.
|
| 2176 | * But it would probably just be a waste of time searching,
|
| 2177 | * or lead to too many vmas hanging off the same anon_vma.
|
| 2178 | * We're trying to allow mprotect remerging later on,
|
| 2179 | * not trying to minimize memory used for anon_vmas.
|
| 2180 |
|
| 2181 | return anon_vma;
|
| 2182 | }
|
| 2183 |
|
| 2184 | static bool vm_ops_needs_writenotify(const struct vm_operations_struct *vm_ops)
|
| 2185 | {
|
| 2186 | return vm_ops && (vm_ops->page_mkwrite || vm_ops->pfn_mkwrite);
|
| 2187 | }
|
| 2188 |
|
| 2189 | static bool vma_is_shared_writable(struct vm_area_struct *vma)
|
| 2190 | {
|
| 2191 | return vma_test_all(vma, VMA_WRITE_BIT, VMA_SHARED_BIT);
|
| 2192 | }
|
| 2193 |
|
| 2194 | static bool vma_fs_can_writeback(struct vm_area_struct *vma)
|
| 2195 | {
|
| 2196 |
|
| 2197 | if (vma_test(vma, VMA_PFNMAP_BIT))
|
| 2198 | return false;
|
| 2199 |
|
| 2200 | return vma->vm_file && vma->vm_file->f_mapping &&
|
| 2201 | mapping_can_writeback(vma->vm_file->f_mapping);
|
| 2202 | }
|
| 2203 |
|
| 2204 | |
| 2205 | * Does this VMA require the underlying folios to have their dirty state
|
| 2206 | * tracked?
|
| 2207 |
|
| 2208 | bool vma_needs_dirty_tracking(struct vm_area_struct *vma)
|
| 2209 | {
|
| 2210 |
|
| 2211 | if (!vma_is_shared_writable(vma))
|
| 2212 | return false;
|
| 2213 |
|
| 2214 |
|
| 2215 | if (vm_ops_needs_writenotify(vma->vm_ops))
|
| 2216 | return true;
|
| 2217 |
|
| 2218 | |
| 2219 | * Even if the filesystem doesn't indicate a need for writenotify, if it
|
| 2220 | * can writeback, dirty tracking is still required.
|
| 2221 |
|
| 2222 | return vma_fs_can_writeback(vma);
|
| 2223 | }
|
| 2224 |
|
| 2225 | |
| 2226 | * Some shared mappings will want the pages marked read-only
|
| 2227 | * to track write events. If so, we'll downgrade vm_page_prot
|
| 2228 | * to the private version (using protection_map[] without the
|
| 2229 | * VM_SHARED bit).
|
| 2230 |
|
| 2231 | bool vma_wants_writenotify(struct vm_area_struct *vma, pgprot_t vm_page_prot)
|
| 2232 | {
|
| 2233 |
|
| 2234 | if (!vma_is_shared_writable(vma))
|
| 2235 | return false;
|
| 2236 |
|
| 2237 |
|
| 2238 | if (vm_ops_needs_writenotify(vma->vm_ops))
|
| 2239 | return true;
|
| 2240 |
|
| 2241 | |
| 2242 | * won't preserve?
|
| 2243 | if (pgprot_val(vm_page_prot) !=
|
| 2244 | pgprot_val(vma_pgprot_modify(vm_page_prot, vma->flags)))
|
| 2245 | return false;
|
| 2246 |
|
| 2247 | |
| 2248 | * Do we need to track softdirty? hugetlb does not support softdirty
|
| 2249 | * tracking yet.
|
| 2250 |
|
| 2251 | if (vma_soft_dirty_enabled(vma) && !is_vm_hugetlb_page(vma))
|
| 2252 | return true;
|
| 2253 |
|
| 2254 |
|
| 2255 | if (userfaultfd_wp(vma))
|
| 2256 | return true;
|
| 2257 |
|
| 2258 |
|
| 2259 | return vma_fs_can_writeback(vma);
|
| 2260 | }
|
| 2261 |
|
| 2262 | static DEFINE_MUTEX(mm_all_locks_mutex);
|
| 2263 |
|
| 2264 | static void vm_lock_anon_vma(struct mm_struct *mm, struct anon_vma *anon_vma)
|
| 2265 | {
|
| 2266 | if (!test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
|
| 2267 | |
| 2268 | * The LSB of head.next can't change from under us
|
| 2269 | * because we hold the mm_all_locks_mutex.
|
| 2270 |
|
| 2271 | down_write_nest_lock(&anon_vma->root->rwsem, &mm->mmap_lock);
|
| 2272 | |
| 2273 | * We can safely modify head.next after taking the
|
| 2274 | * anon_vma->root->rwsem. If some other vma in this mm shares
|
| 2275 | * the same anon_vma we won't take it again.
|
| 2276 | *
|
| 2277 | * No need of atomic instructions here, head.next
|
| 2278 | * can't change from under us thanks to the
|
| 2279 | * anon_vma->root->rwsem.
|
| 2280 |
|
| 2281 | if (__test_and_set_bit(0, (unsigned long *)
|
| 2282 | &anon_vma->root->rb_root.rb_root.rb_node))
|
| 2283 | BUG();
|
| 2284 | }
|
| 2285 | }
|
| 2286 |
|
| 2287 | static void vm_lock_mapping(struct mm_struct *mm, struct address_space *mapping)
|
| 2288 | {
|
| 2289 | if (!test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
|
| 2290 | |
| 2291 | * AS_MM_ALL_LOCKS can't change from under us because
|
| 2292 | * we hold the mm_all_locks_mutex.
|
| 2293 | *
|
| 2294 | * Operations on ->flags have to be atomic because
|
| 2295 | * even if AS_MM_ALL_LOCKS is stable thanks to the
|
| 2296 | * mm_all_locks_mutex, there may be other cpus
|
| 2297 | * changing other bitflags in parallel to us.
|
| 2298 |
|
| 2299 | if (test_and_set_bit(AS_MM_ALL_LOCKS, &mapping->flags))
|
| 2300 | BUG();
|
| 2301 | down_write_nest_lock(&mapping->i_mmap_rwsem, &mm->mmap_lock);
|
| 2302 | }
|
| 2303 | }
|
| 2304 |
|
| 2305 | |
| 2306 | * This operation locks against the VM for all pte/vma/mm related
|
| 2307 | * operations that could ever happen on a certain mm. This includes
|
| 2308 | * vmtruncate, try_to_unmap, and all page faults.
|
| 2309 | *
|
| 2310 | * The caller must take the mmap_lock in write mode before calling
|
| 2311 | * mm_take_all_locks(). The caller isn't allowed to release the
|
| 2312 | * mmap_lock until mm_drop_all_locks() returns.
|
| 2313 | *
|
| 2314 | * mmap_lock in write mode is required in order to block all operations
|
| 2315 | * that could modify pagetables and free pages without need of
|
| 2316 | * altering the vma layout. It's also needed in write mode to avoid new
|
| 2317 | * anon_vmas to be associated with existing vmas.
|
| 2318 | *
|
| 2319 | * A single task can't take more than one mm_take_all_locks() in a row
|
| 2320 | * or it would deadlock.
|
| 2321 | *
|
| 2322 | * The LSB in anon_vma->rb_root.rb_node and the AS_MM_ALL_LOCKS bitflag in
|
| 2323 | * mapping->flags avoid to take the same lock twice, if more than one
|
| 2324 | * vma in this mm is backed by the same anon_vma or address_space.
|
| 2325 | *
|
| 2326 | * We take locks in following order, accordingly to comment at beginning
|
| 2327 | * of mm/rmap.c:
|
| 2328 | * - all hugetlbfs_i_mmap_rwsem_key locks (aka mapping->i_mmap_rwsem for
|
| 2329 | * hugetlb mapping);
|
| 2330 | * - all vmas marked locked
|
| 2331 | * - all i_mmap_rwsem locks;
|
| 2332 | * - all anon_vma->rwseml
|
| 2333 | *
|
| 2334 | * We can take all locks within these types randomly because the VM code
|
| 2335 | * doesn't nest them and we protected from parallel mm_take_all_locks() by
|
| 2336 | * mm_all_locks_mutex.
|
| 2337 | *
|
| 2338 | * mm_take_all_locks() and mm_drop_all_locks are expensive operations
|
| 2339 | * that may have to take thousand of locks.
|
| 2340 | *
|
| 2341 | * mm_take_all_locks() can fail if it's interrupted by signals.
|
| 2342 |
|
| 2343 | int mm_take_all_locks(struct mm_struct *mm)
|
| 2344 | {
|
| 2345 | struct vm_area_struct *vma;
|
| 2346 | struct anon_vma_chain *avc;
|
| 2347 | VMA_ITERATOR(vmi, mm, 0);
|
| 2348 |
|
| 2349 | mmap_assert_write_locked(mm);
|
| 2350 |
|
| 2351 | mutex_lock(&mm_all_locks_mutex);
|
| 2352 |
|
| 2353 | |
| 2354 | * vma_start_write() does not have a complement in mm_drop_all_locks()
|
| 2355 | * because vma_start_write() is always asymmetrical; it marks a VMA as
|
| 2356 | * being written to until mmap_write_unlock() or mmap_write_downgrade()
|
| 2357 | * is reached.
|
| 2358 |
|
| 2359 | for_each_vma(vmi, vma) {
|
| 2360 | if (signal_pending(current))
|
| 2361 | goto out_unlock;
|
| 2362 | vma_start_write(vma);
|
| 2363 | }
|
| 2364 |
|
| 2365 | vma_iter_init(&vmi, mm, 0);
|
| 2366 | for_each_vma(vmi, vma) {
|
| 2367 | if (signal_pending(current))
|
| 2368 | goto out_unlock;
|
| 2369 | if (vma->vm_file && vma->vm_file->f_mapping &&
|
| 2370 | is_vm_hugetlb_page(vma))
|
| 2371 | vm_lock_mapping(mm, vma->vm_file->f_mapping);
|
| 2372 | }
|
| 2373 |
|
| 2374 | vma_iter_init(&vmi, mm, 0);
|
| 2375 | for_each_vma(vmi, vma) {
|
| 2376 | if (signal_pending(current))
|
| 2377 | goto out_unlock;
|
| 2378 | if (vma->vm_file && vma->vm_file->f_mapping &&
|
| 2379 | !is_vm_hugetlb_page(vma))
|
| 2380 | vm_lock_mapping(mm, vma->vm_file->f_mapping);
|
| 2381 | }
|
| 2382 |
|
| 2383 | vma_iter_init(&vmi, mm, 0);
|
| 2384 | for_each_vma(vmi, vma) {
|
| 2385 | if (signal_pending(current))
|
| 2386 | goto out_unlock;
|
| 2387 | if (vma->anon_vma)
|
| 2388 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
|
| 2389 | vm_lock_anon_vma(mm, avc->anon_vma);
|
| 2390 | }
|
| 2391 |
|
| 2392 | return 0;
|
| 2393 |
|
| 2394 | out_unlock:
|
| 2395 | mm_drop_all_locks(mm);
|
| 2396 | return -EINTR;
|
| 2397 | }
|
| 2398 |
|
| 2399 | static void vm_unlock_anon_vma(struct anon_vma *anon_vma)
|
| 2400 | {
|
| 2401 | if (test_bit(0, (unsigned long *) &anon_vma->root->rb_root.rb_root.rb_node)) {
|
| 2402 | |
| 2403 | * The LSB of head.next can't change to 0 from under
|
| 2404 | * us because we hold the mm_all_locks_mutex.
|
| 2405 | *
|
| 2406 | * We must however clear the bitflag before unlocking
|
| 2407 | * the vma so the users using the anon_vma->rb_root will
|
| 2408 | * never see our bitflag.
|
| 2409 | *
|
| 2410 | * No need of atomic instructions here, head.next
|
| 2411 | * can't change from under us until we release the
|
| 2412 | * anon_vma->root->rwsem.
|
| 2413 |
|
| 2414 | if (!__test_and_clear_bit(0, (unsigned long *)
|
| 2415 | &anon_vma->root->rb_root.rb_root.rb_node))
|
| 2416 | BUG();
|
| 2417 | anon_vma_unlock_write(anon_vma);
|
| 2418 | }
|
| 2419 | }
|
| 2420 |
|
| 2421 | static void vm_unlock_mapping(struct address_space *mapping)
|
| 2422 | {
|
| 2423 | if (test_bit(AS_MM_ALL_LOCKS, &mapping->flags)) {
|
| 2424 | |
| 2425 | * AS_MM_ALL_LOCKS can't change to 0 from under us
|
| 2426 | * because we hold the mm_all_locks_mutex.
|
| 2427 |
|
| 2428 | i_mmap_unlock_write(mapping);
|
| 2429 | if (!test_and_clear_bit(AS_MM_ALL_LOCKS,
|
| 2430 | &mapping->flags))
|
| 2431 | BUG();
|
| 2432 | }
|
| 2433 | }
|
| 2434 |
|
| 2435 | |
| 2436 | * The mmap_lock cannot be released by the caller until
|
| 2437 | * mm_drop_all_locks() returns.
|
| 2438 |
|
| 2439 | void mm_drop_all_locks(struct mm_struct *mm)
|
| 2440 | {
|
| 2441 | struct vm_area_struct *vma;
|
| 2442 | struct anon_vma_chain *avc;
|
| 2443 | VMA_ITERATOR(vmi, mm, 0);
|
| 2444 |
|
| 2445 | mmap_assert_write_locked(mm);
|
| 2446 | BUG_ON(!mutex_is_locked(&mm_all_locks_mutex));
|
| 2447 |
|
| 2448 | for_each_vma(vmi, vma) {
|
| 2449 | if (vma->anon_vma)
|
| 2450 | list_for_each_entry(avc, &vma->anon_vma_chain, same_vma)
|
| 2451 | vm_unlock_anon_vma(avc->anon_vma);
|
| 2452 | if (vma->vm_file && vma->vm_file->f_mapping)
|
| 2453 | vm_unlock_mapping(vma->vm_file->f_mapping);
|
| 2454 | }
|
| 2455 |
|
| 2456 | mutex_unlock(&mm_all_locks_mutex);
|
| 2457 | }
|
| 2458 |
|
| 2459 | |
| 2460 | * We account for memory if it's a private writeable mapping,
|
| 2461 | * not hugepages and VM_NORESERVE wasn't set.
|
| 2462 |
|
| 2463 | static bool accountable_mapping(struct mmap_state *map)
|
| 2464 | {
|
| 2465 | const struct file *file = map->file;
|
| 2466 |
|
| 2467 | |
| 2468 | * hugetlb has its own accounting separate from the core VM
|
| 2469 | * VM_HUGETLB may not be set yet so we cannot check for that flag.
|
| 2470 |
|
| 2471 | if (file && is_file_hugepages(file))
|
| 2472 | return false;
|
| 2473 |
|
| 2474 | return vma_flags_test(&map->vma_flags, VMA_WRITE_BIT) &&
|
| 2475 | !vma_flags_test_any(&map->vma_flags, VMA_NORESERVE_BIT,
|
| 2476 | VMA_SHARED_BIT);
|
| 2477 | }
|
| 2478 |
|
| 2479 | |
| 2480 | * vms_abort_munmap_vmas() - Undo as much as possible from an aborted munmap()
|
| 2481 | * operation.
|
| 2482 | * @vms: The vma unmap structure
|
| 2483 | * @mas_detach: The maple state with the detached maple tree
|
| 2484 | *
|
| 2485 | * Reattach any detached vmas, free up the maple tree used to track the vmas.
|
| 2486 | * If that's not possible because the ptes are cleared (and vm_ops->closed() may
|
| 2487 | * have been called), then a NULL is written over the vmas and the vmas are
|
| 2488 | * removed (munmap() completed).
|
| 2489 |
|
| 2490 | static void vms_abort_munmap_vmas(struct vma_munmap_struct *vms,
|
| 2491 | struct ma_state *mas_detach)
|
| 2492 | {
|
| 2493 | struct ma_state *mas = &vms->vmi->mas;
|
| 2494 |
|
| 2495 | if (!vms->nr_pages)
|
| 2496 | return;
|
| 2497 |
|
| 2498 | if (vms->clear_ptes)
|
| 2499 | return reattach_vmas(mas_detach);
|
| 2500 |
|
| 2501 | |
| 2502 | * Aborting cannot just call the vm_ops open() because they are often
|
| 2503 | * not symmetrical and state data has been lost. Resort to the old
|
| 2504 | * failure method of leaving a gap where the MAP_FIXED mapping failed.
|
| 2505 |
|
| 2506 | mas_set_range(mas, vms->start, vms->end - 1);
|
| 2507 | mas_store_gfp(mas, NULL, GFP_KERNEL|__GFP_NOFAIL);
|
| 2508 |
|
| 2509 | vms_complete_munmap_vmas(vms, mas_detach);
|
| 2510 | }
|
| 2511 |
|
| 2512 | static void update_ksm_flags(struct mmap_state *map)
|
| 2513 | {
|
| 2514 | map->vma_flags = ksm_vma_flags(map->mm, map->file, map->vma_flags);
|
| 2515 | }
|
| 2516 |
|
| 2517 | static void set_desc_from_map(struct vm_area_desc *desc,
|
| 2518 | const struct mmap_state *map)
|
| 2519 | {
|
| 2520 | desc->start = map->addr;
|
| 2521 | desc->end = map->end;
|
| 2522 |
|
| 2523 | desc->pgoff = map->pgoff;
|
| 2524 | desc->vm_file = map->file;
|
| 2525 | desc->vma_flags = map->vma_flags;
|
| 2526 | desc->page_prot = map->page_prot;
|
| 2527 | }
|
| 2528 |
|
| 2529 | |
| 2530 | * __mmap_setup() - Prepare to gather any overlapping VMAs that need to be
|
| 2531 | * unmapped once the map operation is completed, check limits, account mapping
|
| 2532 | * and clean up any pre-existing VMAs.
|
| 2533 | *
|
| 2534 | * As a result it sets up the @map and @desc objects.
|
| 2535 | *
|
| 2536 | * @map: Mapping state.
|
| 2537 | * @desc: VMA descriptor
|
| 2538 | * @uf: Userfaultfd context list.
|
| 2539 | *
|
| 2540 | * Returns: 0 on success, error code otherwise.
|
| 2541 |
|
| 2542 | static int __mmap_setup(struct mmap_state *map, struct vm_area_desc *desc,
|
| 2543 | struct list_head *uf)
|
| 2544 | {
|
| 2545 | int error;
|
| 2546 | struct vma_iterator *vmi = map->vmi;
|
| 2547 | struct vma_munmap_struct *vms = &map->vms;
|
| 2548 |
|
| 2549 |
|
| 2550 | vms->vma = vma_find(vmi, map->end);
|
| 2551 | init_vma_munmap(vms, vmi, vms->vma, map->addr, map->end, uf,
|
| 2552 | false);
|
| 2553 |
|
| 2554 |
|
| 2555 | if (vms->vma) {
|
| 2556 | mt_init_flags(&map->mt_detach,
|
| 2557 | vmi->mas.tree->ma_flags & MT_FLAGS_LOCK_MASK);
|
| 2558 | mt_on_stack(map->mt_detach);
|
| 2559 | mas_init(&map->mas_detach, &map->mt_detach, 0);
|
| 2560 |
|
| 2561 | error = vms_gather_munmap_vmas(vms, &map->mas_detach);
|
| 2562 | if (error) {
|
| 2563 |
|
| 2564 | vms->nr_pages = 0;
|
| 2565 | return error;
|
| 2566 | }
|
| 2567 |
|
| 2568 | map->next = vms->next;
|
| 2569 | map->prev = vms->prev;
|
| 2570 | } else {
|
| 2571 | map->next = vma_iter_next_rewind(vmi, &map->prev);
|
| 2572 | }
|
| 2573 |
|
| 2574 |
|
| 2575 | if (!may_expand_vm(map->mm, &map->vma_flags, map->pglen - vms->nr_pages))
|
| 2576 | return -ENOMEM;
|
| 2577 |
|
| 2578 |
|
| 2579 | if (accountable_mapping(map)) {
|
| 2580 | map->charged = map->pglen;
|
| 2581 | map->charged -= vms->nr_accounted;
|
| 2582 | if (map->charged) {
|
| 2583 | error = security_vm_enough_memory_mm(map->mm, map->charged);
|
| 2584 | if (error)
|
| 2585 | return error;
|
| 2586 | }
|
| 2587 |
|
| 2588 | vms->nr_accounted = 0;
|
| 2589 | vma_flags_set(&map->vma_flags, VMA_ACCOUNT_BIT);
|
| 2590 | }
|
| 2591 |
|
| 2592 | |
| 2593 | * Clear PTEs while the vma is still in the tree so that rmap
|
| 2594 | * cannot race with the freeing later in the truncate scenario.
|
| 2595 | * This is also needed for mmap_file(), which is why vm_ops
|
| 2596 | * close function is called.
|
| 2597 |
|
| 2598 | vms_clean_up_area(vms, &map->mas_detach);
|
| 2599 |
|
| 2600 | set_desc_from_map(desc, map);
|
| 2601 | return 0;
|
| 2602 | }
|
| 2603 |
|
| 2604 |
|
| 2605 | static int __mmap_new_file_vma(struct mmap_state *map,
|
| 2606 | struct vm_area_struct *vma)
|
| 2607 | {
|
| 2608 | struct vma_iterator *vmi = map->vmi;
|
| 2609 | int error;
|
| 2610 |
|
| 2611 | vma->vm_file = map->file;
|
| 2612 | if (!map->file_doesnt_need_get)
|
| 2613 | get_file(map->file);
|
| 2614 |
|
| 2615 | if (!map->file->f_op->mmap)
|
| 2616 | return 0;
|
| 2617 |
|
| 2618 | error = mmap_file(vma->vm_file, vma);
|
| 2619 | if (error) {
|
| 2620 | UNMAP_STATE(unmap, vmi, vma, vma->vm_start, vma->vm_end,
|
| 2621 | map->prev, map->next);
|
| 2622 | fput(vma->vm_file);
|
| 2623 | vma->vm_file = NULL;
|
| 2624 |
|
| 2625 | vma_iter_set(vmi, vma->vm_end);
|
| 2626 |
|
| 2627 | unmap_region(&unmap);
|
| 2628 | return error;
|
| 2629 | }
|
| 2630 |
|
| 2631 |
|
| 2632 | WARN_ON_ONCE(map->addr != vma->vm_start);
|
| 2633 | |
| 2634 | * Drivers should not permit writability when previously it was
|
| 2635 | * disallowed.
|
| 2636 |
|
| 2637 | VM_WARN_ON_ONCE(!vma_flags_same_pair(&map->vma_flags, &vma->flags) &&
|
| 2638 | !vma_flags_test(&map->vma_flags, VMA_MAYWRITE_BIT) &&
|
| 2639 | vma_test(vma, VMA_MAYWRITE_BIT));
|
| 2640 |
|
| 2641 | map->file = vma->vm_file;
|
| 2642 | map->vma_flags = vma->flags;
|
| 2643 |
|
| 2644 | return 0;
|
| 2645 | }
|
| 2646 |
|
| 2647 | |
| 2648 | * __mmap_new_vma() - Allocate a new VMA for the region, as merging was not
|
| 2649 | * possible.
|
| 2650 | *
|
| 2651 | * @map: Mapping state.
|
| 2652 | * @vmap: Output pointer for the new VMA.
|
| 2653 | * @action: Any mmap_prepare action that is still to complete.
|
| 2654 | *
|
| 2655 | * Returns: Zero on success, or an error.
|
| 2656 |
|
| 2657 | static int __mmap_new_vma(struct mmap_state *map, struct vm_area_struct **vmap,
|
| 2658 | struct mmap_action *action)
|
| 2659 | {
|
| 2660 | const bool is_anon = !map->file &&
|
| 2661 | !vma_flags_test(&map->vma_flags, VMA_SHARED_BIT);
|
| 2662 | struct vma_iterator *vmi = map->vmi;
|
| 2663 | int error = 0;
|
| 2664 | struct vm_area_struct *vma;
|
| 2665 |
|
| 2666 | |
| 2667 | * Determine the object being mapped and call the appropriate
|
| 2668 | * specific mapper. the address has already been validated, but
|
| 2669 | * not unmapped, but the maps are removed from the list.
|
| 2670 |
|
| 2671 | vma = vm_area_alloc(map->mm);
|
| 2672 | if (!vma)
|
| 2673 | return -ENOMEM;
|
| 2674 |
|
| 2675 | vma_iter_config(vmi, map->addr, map->end);
|
| 2676 |
|
| 2677 | if (is_anon)
|
| 2678 | vma_set_anonymous(vma);
|
| 2679 |
|
| 2680 | vma_set_range(vma, map->addr, map->end, map->pgoff, map->anon_pgoff);
|
| 2681 | vma->flags = map->vma_flags;
|
| 2682 | vma->vm_page_prot = map->page_prot;
|
| 2683 |
|
| 2684 | if (vma_iter_prealloc(vmi, vma)) {
|
| 2685 | error = -ENOMEM;
|
| 2686 | goto free_vma;
|
| 2687 | }
|
| 2688 |
|
| 2689 |
|
| 2690 | if (map->file)
|
| 2691 | error = __mmap_new_file_vma(map, vma);
|
| 2692 | else if (!is_anon)
|
| 2693 | error = shmem_zero_setup(vma);
|
| 2694 |
|
| 2695 | if (error)
|
| 2696 | goto free_iter_vma;
|
| 2697 |
|
| 2698 | if (!map->check_ksm_early) {
|
| 2699 | update_ksm_flags(map);
|
| 2700 | vma->flags = map->vma_flags;
|
| 2701 | }
|
| 2702 |
|
| 2703 | #ifdef CONFIG_SPARC64
|
| 2704 |
|
| 2705 | WARN_ON_ONCE(!arch_validate_flags(map->vm_flags));
|
| 2706 | #endif
|
| 2707 |
|
| 2708 |
|
| 2709 | vma_start_write(vma);
|
| 2710 | vma_iter_store_new(vmi, vma);
|
| 2711 | map->mm->map_count++;
|
| 2712 | vma_link_file(vma, action->hide_from_rmap_until_complete);
|
| 2713 |
|
| 2714 | |
| 2715 | * vma_merge_new_range() calls khugepaged_enter_vma() too, the below
|
| 2716 | * call covers the non-merge case.
|
| 2717 |
|
| 2718 | if (!vma_is_anonymous(vma))
|
| 2719 | khugepaged_enter_vma(vma, map->vm_flags);
|
| 2720 | *vmap = vma;
|
| 2721 | return 0;
|
| 2722 |
|
| 2723 | free_iter_vma:
|
| 2724 | vma_iter_free(vmi);
|
| 2725 | free_vma:
|
| 2726 | vm_area_free(vma);
|
| 2727 | return error;
|
| 2728 | }
|
| 2729 |
|
| 2730 | |
| 2731 | * __mmap_complete() - Unmap any VMAs we overlap, account memory mapping
|
| 2732 | * statistics, handle locking and finalise the VMA.
|
| 2733 | *
|
| 2734 | * @map: Mapping state.
|
| 2735 | * @vma: Merged or newly allocated VMA for the mmap()'d region.
|
| 2736 |
|
| 2737 | static void __mmap_complete(struct mmap_state *map, struct vm_area_struct *vma)
|
| 2738 | {
|
| 2739 | struct mm_struct *mm = map->mm;
|
| 2740 |
|
| 2741 | perf_event_mmap(vma);
|
| 2742 |
|
| 2743 |
|
| 2744 | vms_complete_munmap_vmas(&map->vms, &map->mas_detach);
|
| 2745 |
|
| 2746 | vm_stat_account(mm, vma->vm_flags, map->pglen);
|
| 2747 | if (vma_test(vma, VMA_LOCKED_BIT)) {
|
| 2748 | if (!vma_supports_mlock(vma))
|
| 2749 | vma_clear_flags_mask(vma, VMA_LOCKED_MASK);
|
| 2750 | else
|
| 2751 | mm->locked_vm += map->pglen;
|
| 2752 | }
|
| 2753 |
|
| 2754 | if (vma->vm_file)
|
| 2755 | uprobe_mmap(vma);
|
| 2756 |
|
| 2757 | |
| 2758 | * New (or expanded) vma always get soft dirty status.
|
| 2759 | * Otherwise user-space soft-dirty page tracker won't
|
| 2760 | * be able to distinguish situation when vma area unmapped,
|
| 2761 | * then new mapped in-place (which must be aimed as
|
| 2762 | * a completely new data area).
|
| 2763 |
|
| 2764 | if (pgtable_supports_soft_dirty())
|
| 2765 | vma_set_flags(vma, VMA_SOFTDIRTY_BIT);
|
| 2766 |
|
| 2767 | vma_set_page_prot(vma);
|
| 2768 | }
|
| 2769 |
|
| 2770 | static int call_action_prepare(struct mmap_state *map,
|
| 2771 | struct vm_area_desc *desc)
|
| 2772 | {
|
| 2773 | int err;
|
| 2774 |
|
| 2775 | err = mmap_action_prepare(desc);
|
| 2776 | if (err)
|
| 2777 | return err;
|
| 2778 |
|
| 2779 | return 0;
|
| 2780 | }
|
| 2781 |
|
| 2782 | |
| 2783 | * Invoke the f_op->mmap_prepare() callback for a file-backed mapping that
|
| 2784 | * specifies it.
|
| 2785 | *
|
| 2786 | * This is called prior to any merge attempt, and updates whitelisted fields
|
| 2787 | * that are permitted to be updated by the caller.
|
| 2788 | *
|
| 2789 | * All but user-defined fields will be pre-populated with original values.
|
| 2790 | *
|
| 2791 | * Returns 0 on success, or an error code otherwise.
|
| 2792 |
|
| 2793 | static int call_mmap_prepare(struct mmap_state *map,
|
| 2794 | struct vm_area_desc *desc)
|
| 2795 | {
|
| 2796 | int err;
|
| 2797 |
|
| 2798 |
|
| 2799 | err = vfs_mmap_prepare(map->file, desc);
|
| 2800 | if (err)
|
| 2801 | return err;
|
| 2802 |
|
| 2803 | err = call_action_prepare(map, desc);
|
| 2804 | if (err)
|
| 2805 | return err;
|
| 2806 |
|
| 2807 |
|
| 2808 | map->pgoff = desc->pgoff;
|
| 2809 | if (desc->vm_file != map->file) {
|
| 2810 | map->file_doesnt_need_get = true;
|
| 2811 | map->file = desc->vm_file;
|
| 2812 | }
|
| 2813 | map->vma_flags = desc->vma_flags;
|
| 2814 | map->page_prot = desc->page_prot;
|
| 2815 |
|
| 2816 | map->vm_ops = desc->vm_ops;
|
| 2817 | map->vm_private_data = desc->private_data;
|
| 2818 |
|
| 2819 | return 0;
|
| 2820 | }
|
| 2821 |
|
| 2822 | static void set_vma_user_defined_fields(struct vm_area_struct *vma,
|
| 2823 | struct mmap_state *map)
|
| 2824 | {
|
| 2825 | if (map->vm_ops)
|
| 2826 | vma->vm_ops = map->vm_ops;
|
| 2827 | else
|
| 2828 | vma_set_anonymous(vma);
|
| 2829 | vma->vm_private_data = map->vm_private_data;
|
| 2830 | }
|
| 2831 |
|
| 2832 | |
| 2833 | * Are we guaranteed no driver can change state such as to preclude KSM merging?
|
| 2834 | * If so, let's set the KSM mergeable flag early so we don't break VMA merging.
|
| 2835 |
|
| 2836 | static bool can_set_ksm_flags_early(struct mmap_state *map)
|
| 2837 | {
|
| 2838 | struct file *file = map->file;
|
| 2839 |
|
| 2840 |
|
| 2841 | if (!file)
|
| 2842 | return true;
|
| 2843 |
|
| 2844 | |
| 2845 | * If .mmap_prepare() is specified, then the driver will have already
|
| 2846 | * manipulated state prior to updating KSM flags. So no need to worry
|
| 2847 | * about mmap callbacks modifying VMA flags after the KSM flag has been
|
| 2848 | * updated here, which could otherwise affect KSM eligibility.
|
| 2849 |
|
| 2850 | if (file->f_op->mmap_prepare)
|
| 2851 | return true;
|
| 2852 |
|
| 2853 |
|
| 2854 | if (shmem_file(file))
|
| 2855 | return true;
|
| 2856 |
|
| 2857 |
|
| 2858 | return false;
|
| 2859 | }
|
| 2860 |
|
| 2861 | static unsigned long __mmap_region(struct file *file, unsigned long addr,
|
| 2862 | unsigned long len, vma_flags_t vma_flags,
|
| 2863 | unsigned long pgoff, struct list_head *uf)
|
| 2864 | {
|
| 2865 | struct mm_struct *mm = current->mm;
|
| 2866 | struct vm_area_struct *vma = NULL;
|
| 2867 | const bool have_mmap_prepare = file && file->f_op->mmap_prepare;
|
| 2868 | VMA_ITERATOR(vmi, mm, addr);
|
| 2869 | const pgoff_t anon_pgoff = addr >> PAGE_SHIFT;
|
| 2870 | MMAP_STATE(map, mm, &vmi, addr, len, pgoff, anon_pgoff, vma_flags, file);
|
| 2871 | struct vm_area_desc desc = {
|
| 2872 | .mm = mm,
|
| 2873 | .file = file,
|
| 2874 | .action = {
|
| 2875 | .type = MMAP_NOTHING,
|
| 2876 | },
|
| 2877 | .vm_ops = &vma_dummy_vm_ops,
|
| 2878 | };
|
| 2879 | bool allocated_new = false;
|
| 2880 | int error;
|
| 2881 |
|
| 2882 | map.check_ksm_early = can_set_ksm_flags_early(&map);
|
| 2883 |
|
| 2884 | error = __mmap_setup(&map, &desc, uf);
|
| 2885 | if (error)
|
| 2886 | goto abort_munmap;
|
| 2887 | if (have_mmap_prepare)
|
| 2888 | error = call_mmap_prepare(&map, &desc);
|
| 2889 | if (error)
|
| 2890 | goto unacct_error;
|
| 2891 |
|
| 2892 | if (map.check_ksm_early)
|
| 2893 | update_ksm_flags(&map);
|
| 2894 |
|
| 2895 |
|
| 2896 | if (map.prev || map.next) {
|
| 2897 | VMG_MMAP_STATE(vmg, &map, NULL);
|
| 2898 |
|
| 2899 | vma = vma_merge_new_range(&vmg);
|
| 2900 | }
|
| 2901 |
|
| 2902 |
|
| 2903 | if (!vma) {
|
| 2904 | error = __mmap_new_vma(&map, &vma, &desc.action);
|
| 2905 | if (error)
|
| 2906 | goto unacct_error;
|
| 2907 | allocated_new = true;
|
| 2908 | }
|
| 2909 |
|
| 2910 | if (have_mmap_prepare && allocated_new)
|
| 2911 | set_vma_user_defined_fields(vma, &map);
|
| 2912 |
|
| 2913 | __mmap_complete(&map, vma);
|
| 2914 |
|
| 2915 | if (have_mmap_prepare && allocated_new) {
|
| 2916 | error = mmap_action_complete(vma, &desc.action,
|
| 2917 | false);
|
| 2918 | if (error)
|
| 2919 | return error;
|
| 2920 | }
|
| 2921 |
|
| 2922 | return addr;
|
| 2923 |
|
| 2924 |
|
| 2925 | unacct_error:
|
| 2926 | if (map.charged)
|
| 2927 | vm_unacct_memory(map.charged);
|
| 2928 | abort_munmap:
|
| 2929 | |
| 2930 | * This indicates that .mmap_prepare has set a new file, differing from
|
| 2931 | * desc->vm_file. But since we're aborting the operation, only the
|
| 2932 | * original file will be cleaned up. Ensure we clean up both.
|
| 2933 |
|
| 2934 | if (map.file_doesnt_need_get)
|
| 2935 | fput(map.file);
|
| 2936 | vms_abort_munmap_vmas(&map.vms, &map.mas_detach);
|
| 2937 | return error;
|
| 2938 | }
|
| 2939 |
|
| 2940 | |
| 2941 | * mmap_region() - Actually perform the userland mapping of a VMA into
|
| 2942 | * current->mm with known, aligned and overflow-checked @addr and @len, and
|
| 2943 | * correctly determined VMA flags @vm_flags and page offset @pgoff.
|
| 2944 | *
|
| 2945 | * This is an internal memory management function, and should not be used
|
| 2946 | * directly.
|
| 2947 | *
|
| 2948 | * The caller must write-lock current->mm->mmap_lock.
|
| 2949 | *
|
| 2950 | * @file: If a file-backed mapping, a pointer to the struct file describing the
|
| 2951 | * file to be mapped, otherwise NULL.
|
| 2952 | * @addr: The page-aligned address at which to perform the mapping.
|
| 2953 | * @len: The page-aligned, non-zero, length of the mapping.
|
| 2954 | * @vma_flags: The VMA flags which should be applied to the mapping.
|
| 2955 | * @pgoff: If @file is specified, the page offset into the file, if not then
|
| 2956 | * the virtual page offset in memory of the anonymous mapping.
|
| 2957 | * @uf: Optionally, a pointer to a list head used for tracking userfaultfd unmap
|
| 2958 | * events.
|
| 2959 | *
|
| 2960 | * Returns: Either an error, or the address at which the requested mapping has
|
| 2961 | * been performed.
|
| 2962 |
|
| 2963 | unsigned long mmap_region(struct file *file, unsigned long addr,
|
| 2964 | unsigned long len, vma_flags_t vma_flags,
|
| 2965 | unsigned long pgoff, struct list_head *uf)
|
| 2966 | {
|
| 2967 | unsigned long ret;
|
| 2968 | bool writable_file_mapping = false;
|
| 2969 |
|
| 2970 | mmap_assert_write_locked(current->mm);
|
| 2971 |
|
| 2972 |
|
| 2973 | if (map_deny_write_exec(&vma_flags, &vma_flags))
|
| 2974 | return -EACCES;
|
| 2975 |
|
| 2976 |
|
| 2977 | if (!arch_validate_flags(vma_flags_to_legacy(vma_flags)))
|
| 2978 | return -EINVAL;
|
| 2979 |
|
| 2980 |
|
| 2981 | if (file && is_shared_maywrite(&vma_flags)) {
|
| 2982 | int error = mapping_map_writable(file->f_mapping);
|
| 2983 |
|
| 2984 | if (error)
|
| 2985 | return error;
|
| 2986 | writable_file_mapping = true;
|
| 2987 | }
|
| 2988 |
|
| 2989 | ret = __mmap_region(file, addr, len, vma_flags, pgoff, uf);
|
| 2990 |
|
| 2991 |
|
| 2992 | if (writable_file_mapping)
|
| 2993 | mapping_unmap_writable(file->f_mapping);
|
| 2994 |
|
| 2995 | validate_mm(current->mm);
|
| 2996 | return ret;
|
| 2997 | }
|
| 2998 |
|
| 2999 | |
| 3000 | * do_brk_flags() - Increase the brk vma if the flags match.
|
| 3001 | * @vmi: The vma iterator
|
| 3002 | * @addr: The start address
|
| 3003 | * @len: The length of the increase
|
| 3004 | * @vma: The vma,
|
| 3005 | * @vma_flags: The VMA Flags
|
| 3006 | *
|
| 3007 | * Extend the brk VMA from addr to addr + len. If the VMA is NULL or the flags
|
| 3008 | * do not match then create a new anonymous VMA. Eventually we may be able to
|
| 3009 | * do some brk-specific accounting here.
|
| 3010 | *
|
| 3011 | * Returns: %0 on success, or otherwise an error.
|
| 3012 |
|
| 3013 | int do_brk_flags(struct vma_iterator *vmi, struct vm_area_struct *vma,
|
| 3014 | unsigned long addr, unsigned long len, vma_flags_t vma_flags)
|
| 3015 | {
|
| 3016 | struct mm_struct *mm = current->mm;
|
| 3017 | const pgoff_t pgoff = addr >> PAGE_SHIFT;
|
| 3018 |
|
| 3019 | |
| 3020 | * Check against address space limits by the changed size
|
| 3021 | * Note: This happens *after* clearing old mappings in some code paths.
|
| 3022 |
|
| 3023 | vma_flags_set_mask(&vma_flags, VMA_DATA_DEFAULT_FLAGS);
|
| 3024 | vma_flags_set(&vma_flags, VMA_ACCOUNT_BIT);
|
| 3025 | vma_flags_set_mask(&vma_flags, mm->def_vma_flags);
|
| 3026 |
|
| 3027 | vma_flags = ksm_vma_flags(mm, NULL, vma_flags);
|
| 3028 | if (!may_expand_vm(mm, &vma_flags, len >> PAGE_SHIFT))
|
| 3029 | return -ENOMEM;
|
| 3030 |
|
| 3031 | if (mm->map_count > get_sysctl_max_map_count())
|
| 3032 | return -ENOMEM;
|
| 3033 |
|
| 3034 | if (security_vm_enough_memory_mm(mm, len >> PAGE_SHIFT))
|
| 3035 | return -ENOMEM;
|
| 3036 |
|
| 3037 | |
| 3038 | * Expand the existing vma if possible; Note that singular lists do not
|
| 3039 | * occur after forking, so the expand will only happen on new VMAs.
|
| 3040 |
|
| 3041 | if (vma && vma->vm_end == addr) {
|
| 3042 | VMG_STATE(vmg, mm, vmi, addr, addr + len, vma_flags, pgoff, pgoff);
|
| 3043 |
|
| 3044 | vmg.prev = vma;
|
| 3045 |
|
| 3046 | vmg.just_expand = true;
|
| 3047 |
|
| 3048 | if (vma_merge_new_range(&vmg))
|
| 3049 | goto out;
|
| 3050 | else if (vmg_nomem(&vmg))
|
| 3051 | goto unacct_fail;
|
| 3052 | }
|
| 3053 |
|
| 3054 | if (vma)
|
| 3055 | vma_iter_next_range(vmi);
|
| 3056 |
|
| 3057 | vma = vm_area_alloc(mm);
|
| 3058 | if (!vma)
|
| 3059 | goto unacct_fail;
|
| 3060 |
|
| 3061 | vma_set_anonymous(vma);
|
| 3062 | vma_set_range(vma, addr, addr + len, pgoff, pgoff);
|
| 3063 | vma->flags = vma_flags;
|
| 3064 | vma->vm_page_prot = vm_get_page_prot(vma_flags_to_legacy(vma_flags));
|
| 3065 | vma_start_write(vma);
|
| 3066 | if (vma_iter_store_gfp(vmi, vma, GFP_KERNEL))
|
| 3067 | goto mas_store_fail;
|
| 3068 |
|
| 3069 | mm->map_count++;
|
| 3070 | validate_mm(mm);
|
| 3071 | out:
|
| 3072 | perf_event_mmap(vma);
|
| 3073 | mm->total_vm += len >> PAGE_SHIFT;
|
| 3074 | mm->data_vm += len >> PAGE_SHIFT;
|
| 3075 | if (vma_flags_test(&vma_flags, VMA_LOCKED_BIT))
|
| 3076 | mm->locked_vm += (len >> PAGE_SHIFT);
|
| 3077 | if (pgtable_supports_soft_dirty())
|
| 3078 | vma_set_flags(vma, VMA_SOFTDIRTY_BIT);
|
| 3079 | return 0;
|
| 3080 |
|
| 3081 | mas_store_fail:
|
| 3082 | vm_area_free(vma);
|
| 3083 | unacct_fail:
|
| 3084 | vm_unacct_memory(len >> PAGE_SHIFT);
|
| 3085 | return -ENOMEM;
|
| 3086 | }
|
| 3087 |
|
| 3088 | |
| 3089 | * unmapped_area() - Find an area between the low_limit and the high_limit with
|
| 3090 | * the correct alignment and offset, all from @info. Note: current->mm is used
|
| 3091 | * for the search.
|
| 3092 | *
|
| 3093 | * @info: The unmapped area information including the range [low_limit -
|
| 3094 | * high_limit), the alignment offset and mask.
|
| 3095 | *
|
| 3096 | * Return: A memory address or -ENOMEM.
|
| 3097 |
|
| 3098 | unsigned long unmapped_area(struct vm_unmapped_area_info *info)
|
| 3099 | {
|
| 3100 | unsigned long length, gap;
|
| 3101 | unsigned long low_limit, high_limit;
|
| 3102 | struct vm_area_struct *tmp;
|
| 3103 | VMA_ITERATOR(vmi, current->mm, 0);
|
| 3104 |
|
| 3105 |
|
| 3106 | length = info->length + info->align_mask + info->start_gap;
|
| 3107 | if (length < info->length)
|
| 3108 | return -ENOMEM;
|
| 3109 |
|
| 3110 | low_limit = info->low_limit;
|
| 3111 | if (low_limit < mmap_min_addr)
|
| 3112 | low_limit = mmap_min_addr;
|
| 3113 | high_limit = info->high_limit;
|
| 3114 | retry:
|
| 3115 | if (vma_iter_area_lowest(&vmi, low_limit, high_limit, length))
|
| 3116 | return -ENOMEM;
|
| 3117 |
|
| 3118 | |
| 3119 | * Adjust for the gap first so it doesn't interfere with the later
|
| 3120 | * alignment. The first step is the minimum needed to fulfill the start
|
| 3121 | * gap, the next step is the minimum to align that. It is the minimum
|
| 3122 | * needed to fulfill both.
|
| 3123 |
|
| 3124 | gap = vma_iter_addr(&vmi) + info->start_gap;
|
| 3125 | gap += (info->align_offset - gap) & info->align_mask;
|
| 3126 | tmp = vma_next(&vmi);
|
| 3127 |
|
| 3128 | if (tmp && vma_test_any_mask(tmp, VMA_STARTGAP_FLAGS)) {
|
| 3129 | if (vm_start_gap(tmp) < gap + length - 1) {
|
| 3130 | low_limit = tmp->vm_end;
|
| 3131 | vma_iter_reset(&vmi);
|
| 3132 | goto retry;
|
| 3133 | }
|
| 3134 | } else {
|
| 3135 | tmp = vma_prev(&vmi);
|
| 3136 | if (tmp && vm_end_gap(tmp) > gap) {
|
| 3137 | low_limit = vm_end_gap(tmp);
|
| 3138 | vma_iter_reset(&vmi);
|
| 3139 | goto retry;
|
| 3140 | }
|
| 3141 | }
|
| 3142 |
|
| 3143 | return gap;
|
| 3144 | }
|
| 3145 |
|
| 3146 | |
| 3147 | * unmapped_area_topdown() - Find an area between the low_limit and the
|
| 3148 | * high_limit with the correct alignment and offset at the highest available
|
| 3149 | * address, all from @info. Note: current->mm is used for the search.
|
| 3150 | *
|
| 3151 | * @info: The unmapped area information including the range [low_limit -
|
| 3152 | * high_limit), the alignment offset and mask.
|
| 3153 | *
|
| 3154 | * Return: A memory address or -ENOMEM.
|
| 3155 |
|
| 3156 | unsigned long unmapped_area_topdown(struct vm_unmapped_area_info *info)
|
| 3157 | {
|
| 3158 | unsigned long length, gap, gap_end;
|
| 3159 | unsigned long low_limit, high_limit;
|
| 3160 | struct vm_area_struct *tmp;
|
| 3161 | VMA_ITERATOR(vmi, current->mm, 0);
|
| 3162 |
|
| 3163 |
|
| 3164 | length = info->length + info->align_mask + info->start_gap;
|
| 3165 | if (length < info->length)
|
| 3166 | return -ENOMEM;
|
| 3167 |
|
| 3168 | low_limit = info->low_limit;
|
| 3169 | if (low_limit < mmap_min_addr)
|
| 3170 | low_limit = mmap_min_addr;
|
| 3171 | high_limit = info->high_limit;
|
| 3172 | retry:
|
| 3173 | if (vma_iter_area_highest(&vmi, low_limit, high_limit, length))
|
| 3174 | return -ENOMEM;
|
| 3175 |
|
| 3176 | gap = vma_iter_end(&vmi) - info->length;
|
| 3177 | gap -= (gap - info->align_offset) & info->align_mask;
|
| 3178 | gap_end = vma_iter_end(&vmi);
|
| 3179 | tmp = vma_next(&vmi);
|
| 3180 |
|
| 3181 | if (tmp && vma_test_any_mask(tmp, VMA_STARTGAP_FLAGS)) {
|
| 3182 | if (vm_start_gap(tmp) < gap_end) {
|
| 3183 | high_limit = vm_start_gap(tmp);
|
| 3184 | vma_iter_reset(&vmi);
|
| 3185 | goto retry;
|
| 3186 | }
|
| 3187 | } else {
|
| 3188 | tmp = vma_prev(&vmi);
|
| 3189 | if (tmp && vm_end_gap(tmp) > gap) {
|
| 3190 | high_limit = tmp->vm_start;
|
| 3191 | vma_iter_reset(&vmi);
|
| 3192 | goto retry;
|
| 3193 | }
|
| 3194 | }
|
| 3195 |
|
| 3196 | return gap;
|
| 3197 | }
|
| 3198 |
|
| 3199 | |
| 3200 | * Verify that the stack growth is acceptable and
|
| 3201 | * update accounting. This is shared with both the
|
| 3202 | * grow-up and grow-down cases.
|
| 3203 |
|
| 3204 | static int acct_stack_growth(struct vm_area_struct *vma,
|
| 3205 | unsigned long size, unsigned long grow)
|
| 3206 | {
|
| 3207 | struct mm_struct *mm = vma->vm_mm;
|
| 3208 | unsigned long new_start;
|
| 3209 |
|
| 3210 |
|
| 3211 | if (!may_expand_vm(mm, &vma->flags, grow))
|
| 3212 | return -ENOMEM;
|
| 3213 |
|
| 3214 |
|
| 3215 | if (size > rlimit(RLIMIT_STACK))
|
| 3216 | return -ENOMEM;
|
| 3217 |
|
| 3218 |
|
| 3219 | if (!mlock_future_ok(mm, vma_test(vma, VMA_LOCKED_BIT),
|
| 3220 | grow << PAGE_SHIFT))
|
| 3221 | return -ENOMEM;
|
| 3222 |
|
| 3223 |
|
| 3224 | new_start = vma->vm_end - size;
|
| 3225 | #ifdef CONFIG_STACK_GROWSUP
|
| 3226 | if (vma_test(vma, VMA_GROWSUP_BIT))
|
| 3227 | new_start = vma->vm_start;
|
| 3228 | #endif
|
| 3229 | if (is_hugepage_only_range(vma->vm_mm, new_start, size))
|
| 3230 | return -EFAULT;
|
| 3231 |
|
| 3232 | |
| 3233 | * Overcommit.. This must be the final test, as it will
|
| 3234 | * update security statistics.
|
| 3235 |
|
| 3236 | if (security_vm_enough_memory_mm(mm, grow))
|
| 3237 | return -ENOMEM;
|
| 3238 |
|
| 3239 | return 0;
|
| 3240 | }
|
| 3241 |
|
| 3242 | #ifdef CONFIG_STACK_GROWSUP
|
| 3243 | |
| 3244 | * PA-RISC uses this for its stack.
|
| 3245 | * vma is the last one with address > vma->vm_end. Have to extend vma.
|
| 3246 |
|
| 3247 | int expand_upwards(struct vm_area_struct *vma, unsigned long address)
|
| 3248 | {
|
| 3249 | struct mm_struct *mm = vma->vm_mm;
|
| 3250 | struct vm_area_struct *next;
|
| 3251 | unsigned long gap_addr;
|
| 3252 | int error = 0;
|
| 3253 | VMA_ITERATOR(vmi, mm, vma->vm_start);
|
| 3254 |
|
| 3255 | if (!vma_test(vma, VMA_GROWSUP_BIT))
|
| 3256 | return -EFAULT;
|
| 3257 |
|
| 3258 | mmap_assert_write_locked(mm);
|
| 3259 |
|
| 3260 |
|
| 3261 | address &= PAGE_MASK;
|
| 3262 | if (address >= (TASK_SIZE & PAGE_MASK))
|
| 3263 | return -ENOMEM;
|
| 3264 | address += PAGE_SIZE;
|
| 3265 |
|
| 3266 |
|
| 3267 | gap_addr = address + stack_guard_gap;
|
| 3268 |
|
| 3269 |
|
| 3270 | if (gap_addr < address || gap_addr > TASK_SIZE)
|
| 3271 | gap_addr = TASK_SIZE;
|
| 3272 |
|
| 3273 | next = find_vma_intersection(mm, vma->vm_end, gap_addr);
|
| 3274 | if (next && vma_is_accessible(next)) {
|
| 3275 | if (!vma_test(next, VMA_GROWSUP_BIT))
|
| 3276 | return -ENOMEM;
|
| 3277 |
|
| 3278 | }
|
| 3279 |
|
| 3280 | if (next)
|
| 3281 | vma_iter_prev_range_limit(&vmi, address);
|
| 3282 |
|
| 3283 | vma_iter_config(&vmi, vma->vm_start, address);
|
| 3284 | if (vma_iter_prealloc(&vmi, vma))
|
| 3285 | return -ENOMEM;
|
| 3286 |
|
| 3287 |
|
| 3288 | if (unlikely(anon_vma_prepare(vma))) {
|
| 3289 | vma_iter_free(&vmi);
|
| 3290 | return -ENOMEM;
|
| 3291 | }
|
| 3292 |
|
| 3293 |
|
| 3294 | vma_start_write(vma);
|
| 3295 |
|
| 3296 | anon_vma_lock_write(vma->anon_vma);
|
| 3297 |
|
| 3298 |
|
| 3299 | if (address > vma->vm_end) {
|
| 3300 | const unsigned long size = address - vma->vm_start;
|
| 3301 | const unsigned long grow = (address - vma->vm_end) >> PAGE_SHIFT;
|
| 3302 | const pgoff_t pgoff = vma_start_pgoff(vma);
|
| 3303 |
|
| 3304 | error = -ENOMEM;
|
| 3305 | if (pgoff + (size >> PAGE_SHIFT) >= pgoff) {
|
| 3306 | error = acct_stack_growth(vma, size, grow);
|
| 3307 | if (!error) {
|
| 3308 | if (vma_test(vma, VMA_LOCKED_BIT))
|
| 3309 | mm->locked_vm += grow;
|
| 3310 | vm_stat_account(mm, vma->vm_flags, grow);
|
| 3311 | anon_rmap_tree_pre_update_vma(vma);
|
| 3312 | vma->vm_end = address;
|
| 3313 |
|
| 3314 | vma_iter_store_overwrite(&vmi, vma);
|
| 3315 | anon_rmap_tree_post_update_vma(vma);
|
| 3316 |
|
| 3317 | perf_event_mmap(vma);
|
| 3318 | }
|
| 3319 | }
|
| 3320 | }
|
| 3321 | anon_vma_unlock_write(vma->anon_vma);
|
| 3322 | vma_iter_free(&vmi);
|
| 3323 | validate_mm(mm);
|
| 3324 | return error;
|
| 3325 | }
|
| 3326 | #endif
|
| 3327 |
|
| 3328 | |
| 3329 | * vma is the first one with address < vma->vm_start. Have to extend vma.
|
| 3330 | * mmap_lock held for writing.
|
| 3331 |
|
| 3332 | int expand_downwards(struct vm_area_struct *vma, unsigned long address)
|
| 3333 | {
|
| 3334 | struct mm_struct *mm = vma->vm_mm;
|
| 3335 | struct vm_area_struct *prev;
|
| 3336 | int error = 0;
|
| 3337 | VMA_ITERATOR(vmi, mm, vma->vm_start);
|
| 3338 |
|
| 3339 | if (!vma_test(vma, VMA_GROWSDOWN_BIT))
|
| 3340 | return -EFAULT;
|
| 3341 |
|
| 3342 | mmap_assert_write_locked(mm);
|
| 3343 |
|
| 3344 | address &= PAGE_MASK;
|
| 3345 | if (address < mmap_min_addr || address < FIRST_USER_ADDRESS)
|
| 3346 | return -EPERM;
|
| 3347 |
|
| 3348 |
|
| 3349 | prev = vma_prev(&vmi);
|
| 3350 |
|
| 3351 | if (prev) {
|
| 3352 | if (!vma_test(prev, VMA_GROWSDOWN_BIT) &&
|
| 3353 | vma_is_accessible(prev) &&
|
| 3354 | (address - prev->vm_end < stack_guard_gap))
|
| 3355 | return -ENOMEM;
|
| 3356 | }
|
| 3357 |
|
| 3358 | if (prev)
|
| 3359 | vma_iter_next_range_limit(&vmi, vma->vm_start);
|
| 3360 |
|
| 3361 | vma_iter_config(&vmi, address, vma->vm_end);
|
| 3362 | if (vma_iter_prealloc(&vmi, vma))
|
| 3363 | return -ENOMEM;
|
| 3364 |
|
| 3365 |
|
| 3366 | if (unlikely(anon_vma_prepare(vma))) {
|
| 3367 | vma_iter_free(&vmi);
|
| 3368 | return -ENOMEM;
|
| 3369 | }
|
| 3370 |
|
| 3371 |
|
| 3372 | vma_start_write(vma);
|
| 3373 |
|
| 3374 | anon_vma_lock_write(vma->anon_vma);
|
| 3375 |
|
| 3376 |
|
| 3377 | if (address < vma->vm_start) {
|
| 3378 | const unsigned long size = vma->vm_end - address;
|
| 3379 | const unsigned long grow = (vma->vm_start - address) >> PAGE_SHIFT;
|
| 3380 |
|
| 3381 | error = -ENOMEM;
|
| 3382 | if (grow <= vma_start_pgoff(vma)) {
|
| 3383 | error = acct_stack_growth(vma, size, grow);
|
| 3384 | if (!error) {
|
| 3385 | if (vma_test(vma, VMA_LOCKED_BIT))
|
| 3386 | mm->locked_vm += grow;
|
| 3387 | vm_stat_account(mm, vma->vm_flags, grow);
|
| 3388 | anon_rmap_tree_pre_update_vma(vma);
|
| 3389 | vma->vm_start = address;
|
| 3390 | vma_sub_pgoff(vma, grow);
|
| 3391 |
|
| 3392 | vma_iter_store_overwrite(&vmi, vma);
|
| 3393 | anon_rmap_tree_post_update_vma(vma);
|
| 3394 |
|
| 3395 | perf_event_mmap(vma);
|
| 3396 | }
|
| 3397 | }
|
| 3398 | }
|
| 3399 | anon_vma_unlock_write(vma->anon_vma);
|
| 3400 | vma_iter_free(&vmi);
|
| 3401 | validate_mm(mm);
|
| 3402 | return error;
|
| 3403 | }
|
| 3404 |
|
| 3405 | int __vm_munmap(unsigned long start, size_t len, bool unlock)
|
| 3406 | {
|
| 3407 | int ret;
|
| 3408 | struct mm_struct *mm = current->mm;
|
| 3409 | LIST_HEAD(uf);
|
| 3410 | VMA_ITERATOR(vmi, mm, start);
|
| 3411 |
|
| 3412 | if (mmap_write_lock_killable(mm))
|
| 3413 | return -EINTR;
|
| 3414 |
|
| 3415 | ret = do_vmi_munmap(&vmi, mm, start, len, &uf, unlock);
|
| 3416 | if (ret || !unlock)
|
| 3417 | mmap_write_unlock(mm);
|
| 3418 |
|
| 3419 | userfaultfd_unmap_complete(mm, &uf);
|
| 3420 | return ret;
|
| 3421 | }
|
| 3422 |
|
| 3423 | |
| 3424 | * Insert vm structure into process list sorted by address
|
| 3425 | * and into the inode's i_mmap tree if file-backed.
|
| 3426 |
|
| 3427 | int insert_vm_struct(struct mm_struct *mm, struct vm_area_struct *vma)
|
| 3428 | {
|
| 3429 | unsigned long charged = vma_pages(vma);
|
| 3430 |
|
| 3431 | if (find_vma_intersection(mm, vma->vm_start, vma->vm_end))
|
| 3432 | return -ENOMEM;
|
| 3433 |
|
| 3434 | if (vma_test(vma, VMA_ACCOUNT_BIT) &&
|
| 3435 | security_vm_enough_memory_mm(mm, charged))
|
| 3436 | return -ENOMEM;
|
| 3437 |
|
| 3438 | |
| 3439 | * The vm_pgoff of a purely anonymous vma should be irrelevant
|
| 3440 | * until its first write fault, when page's anon_vma and index
|
| 3441 | * are set. But now set the vm_pgoff it will almost certainly
|
| 3442 | * end up with (unless mremap moves it elsewhere before that
|
| 3443 | * first wfault), so /proc/pid/maps tells a consistent story.
|
| 3444 | *
|
| 3445 | * By setting it to reflect the virtual start address of the
|
| 3446 | * vma, merges and splits can happen in a seamless way, just
|
| 3447 | * using the existing file pgoff checks and manipulations.
|
| 3448 | * Similarly in do_mmap and in do_brk_flags.
|
| 3449 |
|
| 3450 | if (vma_is_anonymous(vma)) {
|
| 3451 | WARN_ON_ONCE(vma->anon_vma);
|
| 3452 | vma_set_pgoff(vma, vma->vm_start >> PAGE_SHIFT);
|
| 3453 | }
|
| 3454 | vma_set_anon_pgoff(vma, vma->vm_start >> PAGE_SHIFT);
|
| 3455 |
|
| 3456 | if (vma_link(mm, vma)) {
|
| 3457 | if (vma_test(vma, VMA_ACCOUNT_BIT))
|
| 3458 | vm_unacct_memory(charged);
|
| 3459 | return -ENOMEM;
|
| 3460 | }
|
| 3461 |
|
| 3462 | return 0;
|
| 3463 | }
|
| 3464 |
|
| 3465 | |
| 3466 | * vma_mmu_pagesize - Default MMU page size granularity for this VMA.
|
| 3467 | * @vma: The user mapping.
|
| 3468 | *
|
| 3469 | * In the common case, the default page size used by the MMU matches the
|
| 3470 | * default page size used by the kernel (see vma_kernel_pagesize()). On
|
| 3471 | * architectures where it differs, an architecture-specific 'strong' version
|
| 3472 | * of this symbol is required.
|
| 3473 | *
|
| 3474 | * The default MMU page size is not affected by Transparent Huge Pages
|
| 3475 | * being in effect, or any usage of larger MMU page sizes (either through
|
| 3476 | * architectural huge-page mappings or other explicit/implicit coalescing of
|
| 3477 | * virtual ranges performed by the MMU).
|
| 3478 | *
|
| 3479 | * Return: The default MMU page size granularity for this VMA.
|
| 3480 |
|
| 3481 | __weak unsigned long vma_mmu_pagesize(struct vm_area_struct *vma)
|
| 3482 | {
|
| 3483 | return vma_kernel_pagesize(vma);
|
| 3484 | }
|
| 3485 |
|
| 3486 | struct vm_area_struct *__install_special_mapping(
|
| 3487 | struct mm_struct *mm,
|
| 3488 | unsigned long addr, unsigned long len,
|
| 3489 | vm_flags_t vm_flags, void *priv,
|
| 3490 | const struct vm_operations_struct *ops)
|
| 3491 | {
|
| 3492 | vma_flags_t vma_flags = legacy_to_vma_flags(vm_flags);
|
| 3493 | struct vm_area_struct *vma;
|
| 3494 | int ret;
|
| 3495 |
|
| 3496 | vma = vm_area_alloc(mm);
|
| 3497 | if (unlikely(!vma))
|
| 3498 | return ERR_PTR(-ENOMEM);
|
| 3499 |
|
| 3500 | vma_flags_set_mask(&vma_flags, mm->def_vma_flags);
|
| 3501 | vma_flags_set(&vma_flags, VMA_DONTEXPAND_BIT);
|
| 3502 | if (pgtable_supports_soft_dirty())
|
| 3503 | vma_flags_set(&vma_flags, VMA_SOFTDIRTY_BIT);
|
| 3504 | vma_flags_clear_mask(&vma_flags, VMA_LOCKED_MASK);
|
| 3505 | vma->flags = vma_flags;
|
| 3506 | vma->vm_page_prot = vma_get_page_prot(vma);
|
| 3507 |
|
| 3508 | vma->vm_ops = ops;
|
| 3509 | vma->vm_private_data = priv;
|
| 3510 | vma_set_range(vma, addr, addr + len, 0, addr >> PAGE_SHIFT);
|
| 3511 |
|
| 3512 | ret = insert_vm_struct(mm, vma);
|
| 3513 | if (ret)
|
| 3514 | goto out;
|
| 3515 |
|
| 3516 | vm_stat_account(mm, vma->vm_flags, len >> PAGE_SHIFT);
|
| 3517 |
|
| 3518 | perf_event_mmap(vma);
|
| 3519 |
|
| 3520 | return vma;
|
| 3521 |
|
| 3522 | out:
|
| 3523 | vm_area_free(vma);
|
| 3524 | return ERR_PTR(ret);
|
| 3525 | }
|